ecole doctorale edite thèse présentée pour l ... · ecole doctorale edite thèse présentée...

Thèse n°

Ecole Doctorale EDITE

Thèse présentée pour l’obtention du diplôme deDocteur de Télécom & Management SudParis

Doctorat conjoint TMSP-UPMC

Spécialité :Informatique, Télécommunications et Electronique

ParTeck Aguilar

TitreVers un protocole de routage géographique avec contention et communications

coopératives pour les réseaux de capteurs

Soutenu le 15 décembre 2010 devant le jury compose de:

André-Luc Beylot Rapporteur INPT/ENSEEIHTLuis Muñoz Rapporteur Université de CantabriaGuy Pujolle Examinateur Université Pierre et Marie CuriePascale Minet Examinateur INRIA Rocquencourt Dominique Barth Examinateur Université de VersaillesHassnaa Moustafa Examinateur Orange Labs, FranceVincent Gauthier Examinateur Telecom Sud ParisHossam Afifi Directeur de

thèseTelecom Sud Paris

Thèse n°

Ecole Doctorale EDITE

Submitted in partial satisfaction of the requirements for the degree of Doctor of sciences in Télécom & Management SudParis

Doctor of Sciences TMSP-UPMC

Specialization:Computer sciences

Presented byTeck Aguilar

TitleToward a Beaconless Geographic Routing with Cooperative Communications for

Wireless Sensor Networks

December 15 2010; committee in charge:

André-Luc Beylot Rapporteur INPT/ENSEEIHTLuis Muñoz Rapporteur Universidad de CantabriaGuy Pujolle Examinateur Université Pierre et Marie CuriePascale Minet Examinateur INRIA Rocquencourt Dominique Barth Examinateur Université de VersaillesHassnaa Moustafa Examinateur Orange Labs, FranceVincent Gauthier Examinateur Telecom Sud ParisHossam Afifi Directeur de

thèseTelecom Sud Paris

Acknowledgement

I would like to acknowledge my advisor Prof. Hossam Afifi who gave me theopportunity to join the Wireless Networks and Multimedia Services Department(RS2M) at Telecom SudParis. I also thank Dr. Vincent Gauthier who so kindlyco-advised this work, spent so many hours helping me and had influenced myresearch. I am deeply grateful for his support.

I thank Syue-Ju, all the friends from Telecom SudParis: Sepideh, Ines, Anahita,Bastien, Makhlouf, Mehdi, Vincent, Abid, Boutabia, Ehmad, Cuiting, Mariem,and so many other that i knew during my research period.

I thank my brothers and sisters and in particular my parents Ronay andSandra - You did a great job!. To my family-in-law for their support and mademe feel at home in a foreign place.

I would like to thank my son Gael and finally to my lovely wife Mariannewho supported and motivated me to continue even in the difficult and frustratingtimes. I love you.

5

To Marianne, Gael, Mateo

and

My parents

Abstract

In Wireless Sensor Networks, the routing task is an essential service thatforwards the sensor readings to some data collection points in the network onthe basis of the multi-hop relaying. The routing task is particularly challengingas it should be realized in an energy efficiency manner with limited amount ofinformation. Geographic routing is a promising approach because of its goodscalability and local information use, but when deploying such approach, someproblems still remain because of some practical difficulties.

In this thesis, some techniques have been explored to address two issues ingeographic routing protocols: i) Cost associated to: the wireless channel impair-ments due to fading, mobility patterns or high dynamic environment and ii) themanagement of constrained resources of the nodes. To tackle these issues, twoprotocols were presented: a beaconless Cooperative Geographic cross-layer proto-col for ad hoc and sensor networks (CoopGeo) and a Relay-Aware CooperativeRouting protocol (RACR).

Unlike traditional geographic routing protocols, CoopGeo deals the wirelessimpairments by means of a cross-layer framework where a beaconless geographicrouting approach was used to build the route not only in a local manner, butalso on the fly worked with a relay selection mechanism to exploit the broadcastnature of the wireless communications.

The RACR protocol exploits the coverage extension as a result from nodecooperation to improve the non-cooperative geographic routing. It is an alterna-tive to scenarios where network resources like energy should be preserved whilerespecting a Symbol Error Rate constraint (SER). Thus, the proposed routingprotocol, enables a node to make a local route decision depending on the geo-graphic location of a relay while this relay is selected with the purpose of providingthe maximum coverage extension toward the destination.

The results obtained from extensive evaluations of CoopGeo and RACR con-tributions, have demonstrated that both solutions are applicable to sensor net-works with very variable channel environments or unpredictable changes in the

9

10

network topology. Therefore, we have proved that our cross-layer vision of theproblem provided an integrated solution to problems like inefficient routing paths,congested medium access, inaccurate location information, and lossy links.

Resume

Le routage dans les reseaux de capteurs, est un service essentiel qui transmetles lectures des capteurs a certains points de collecte de donnees dans le reseausur la base des relais multi-saut. Cette tache est particulierement difficile car elledoit etre realise d’une maniere efficace au niveau de consommation de resourceset avec une quantite limitee d’informations disponible. La facilite de mise al’echelle et l’utilisation d’information local pour fonctionner ont permis au routagegeographique etre considere comme une approche prometteuse. Cependant, lorsde son implementation, certains problemes subsistent en raison des difficultespratiques.

Dans ce travail de recherche, two problematiques inherentes aux protocoles deroutages geographique ont ete etudies: i) Le cout associe: aux evanouissementslies aux obstacles et aux multi-trajets suivis par un signal transmis sur un canalradio, aux changements rapides des conditions physiques du canal de transmissionand ii) l’administration de resources affectes a chaque noeud appartenant aureseau. Afin de resoudre ce probleme, deux protocoles ont ete presentes: unprotocole de routage geographique avec communications cooperatives, beaconlessCooperative cooperative Geographic cross-layer protocol for ad hoc and sensornetworks (CoopGeo) et un protocole de routage base sur le principe d’extensionde couverture Relay-Aware Cooperative Routing (RACR).

Contrairement aux protocoles de routage geographique traditionnelles, Coop-Geo est un protocole de routage ”beaconless” base sur une architecture inter-couches ou le routage non seulement est realise localement, mais aussi a la vole.En plus, les problemes lies a la couche physique sont traites par les communica-tions cooperatives qui exploitaient la nature de la diffusion sans fil.

Le protocole RACR exploite la propriete offert par les communications cooperatives:l’extension de la couverture radio. Cet propriete permets d’ameliorer les perfor-mances d’un reseau que a l’origine utilise un protocole de routage geographiquetraditionnel. RACR est une alternative aux scenarios ou l’objectif principal estcelui de diminuer au maximum la consommation des resources du reseau et au

11

12

meme temps assurer que le reseau offre un taux d’erreur par symbole garanti(SER). Ainsi, le protocole RACR, permets a un noeud effectuer des decisionsdites locales, par rapport au routage des paquets qui dependent de la localisationgeographique d’un noeud relai, tandis que, ce noued relai a la finalite de donnerune extension maximale au niveau de couverture radio envers la destination.

Les resultats obtenus a partir des evaluations approfondies de CoopGeo etRACR, ont demontre que les deux solutions sont applicables aux reseaux de cap-teurs en presence forte mobilite, environnements tres variables au niveau radio, ouavec des erreurs aux niveau de l’information de localisation. Par consequent, nousavons prouve que notre vision de inter-couche du probleme a fourni deux solutionsefficaces, en termes de chemins, acces au media, problemes lies a l’informationimprecise de localisation, et des liens perturbes.

Contents

1 Introduction 171.1 Context and challenges . . . . . . . . . . . . . . . . . . . . . . . . 181.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201.3 Thesis Outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21

2 State of the art 232.1 Routing Protocols for Wireless Sensor Networks . . . . . . . . . . 23

2.1.1 Topology-based routing . . . . . . . . . . . . . . . . . . . . 242.1.2 Position-based routing (Geographic Routing) . . . . . . . . 25

2.2 Geographic Routing Protocols . . . . . . . . . . . . . . . . . . . . 252.2.1 Forwarding Strategies ”Greedy Forwarding” . . . . . . . . 272.2.2 Some Traditional Geographic routing protocols . . . . . . 29

2.2.2.1 Greedy Face Greedy (GFG) . . . . . . . . . . . . 292.2.2.2 Greedy Perimeter Stateless Routing (GPSR) . . . 292.2.2.3 Other Geographic protocols derived from greedy

+ face union . . . . . . . . . . . . . . . . . . . . 302.2.2.4 Greedy Distributed Spanning Tree Routing (GDSTR) 302.2.2.5 Local Tree based Greedy Routing (LTGR) . . . . 31

2.2.3 Beaconless geographic routing (BLGR) . . . . . . . . . . . 312.2.3.1 Beaconless Routing (BLR) . . . . . . . . . . . . . 332.2.3.2 Contention-Based Forwarding (CBF) . . . . . . . 342.2.3.3 Implicit Geographic Forwarding (IGF) . . . . . . 352.2.3.4 Geographic Random Forwarding (GeRaF) . . . . 36

2.2.4 New Beaconless Geographic Routing Protocols . . . . . . . 372.2.4.1 Beaconless On Demand Strategy for Geographic

Routing in Wireless Sensor Networks (BOSS) . . 372.2.4.2 Select and Protest-based Beaconless Georouting

with guaranteed delivery . . . . . . . . . . . . . . 382.3 Virtual Coordinates . . . . . . . . . . . . . . . . . . . . . . . . . . 39

13

14 Contents

2.3.1 Location-aware landmarks . . . . . . . . . . . . . . . . . . 392.3.2 Location-unaware landmarks . . . . . . . . . . . . . . . . . 402.3.3 Landmark-free Virtual Coordinates . . . . . . . . . . . . . 42

2.4 Cooperative Communications . . . . . . . . . . . . . . . . . . . . 422.4.1 General concept . . . . . . . . . . . . . . . . . . . . . . . . 432.4.2 Relay behavior and protocols . . . . . . . . . . . . . . . . 432.4.3 Cooperative communications classification . . . . . . . . . 442.4.4 Relay selection . . . . . . . . . . . . . . . . . . . . . . . . 462.4.5 Single relay selection . . . . . . . . . . . . . . . . . . . . . 47

2.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49

3 Models and tools 513.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 513.2 Network models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52

3.2.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 563.3 Power-attenuation model . . . . . . . . . . . . . . . . . . . . . . . 563.4 Radio propagation models . . . . . . . . . . . . . . . . . . . . . . 57

3.4.1 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.5 Mobility models . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593.6 Simulation tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

4 CoopGeo: A Cooperative Geographic Routing Protocol 634.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 644.2 Network Model and Problem Statement . . . . . . . . . . . . . . . 67

4.2.1 Network Model . . . . . . . . . . . . . . . . . . . . . . . . 674.2.2 Problem Statement . . . . . . . . . . . . . . . . . . . . . . 70

4.3 CoopGeo: A geographic cross-layer protocol for cooperative wire-less networks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 724.3.1 Beaconless Greedy Forwarding (BLGF) . . . . . . . . . . . 72

4.3.1.1 Geographic contention-based forwarder selection(TCBF ) . . . . . . . . . . . . . . . . . . . . . . . 73

4.3.2 Beaconless Recovery Forwarding (BLRF) . . . . . . . . . . 734.3.3 MAC-PHY Cross-Layered Relay Selection . . . . . . . . . 76

4.3.3.1 Relay selection criterion based on geographical in-formation . . . . . . . . . . . . . . . . . . . . . . 76

4.3.3.2 Geographic contention-based relay selection . . . 774.3.3.3 Relay selection area . . . . . . . . . . . . . . . . 78

4.3.4 CoopGeo in Action . . . . . . . . . . . . . . . . . . . . . . 794.4 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . 81

4.4.1 Packet Error Rate (PER) . . . . . . . . . . . . . . . . . . 844.4.2 End to End Transmission Error Probability . . . . . . . . 85

Contents 15

4.4.3 Varying the contention window Tmax . . . . . . . . . . . . 854.4.3.1 CTF-Relayed message Collision Probability . . . 854.4.3.2 Varying the constellation size . . . . . . . . . . . 86

4.5 Discussion and conclusions . . . . . . . . . . . . . . . . . . . . . . 87

5 RACR: Relay-Aware Cooperative Routing 915.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 915.2 System Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

5.2.1 Signal Model . . . . . . . . . . . . . . . . . . . . . . . . . 935.2.1.1 Cooperative Transmission . . . . . . . . . . . . . 935.2.1.2 Direct Transmission . . . . . . . . . . . . . . . . 94

5.2.2 Channel Model . . . . . . . . . . . . . . . . . . . . . . . . 945.2.3 Theoretical Average SER Performances . . . . . . . . . . . 95

5.2.3.1 Average SER under Direct Transmission . . . . . 955.2.3.2 Average SER under DF Cooperative Transmission 95

5.2.4 Network Model . . . . . . . . . . . . . . . . . . . . . . . . 965.3 SER-Based Radio Coverage Formulation . . . . . . . . . . . . . . 975.4 RACR: Relay-Aware Cooperative Routing . . . . . . . . . . . . . 100

5.4.1 Contention Timer Setting for Relay Selection . . . . . . . 1015.4.2 Contention Timer Setting for Forwarder Selection . . . . . 102

5.5 Performance Evaluation . . . . . . . . . . . . . . . . . . . . . . . 1035.5.1 Should we choose the relay as far as possible from the source?1035.5.2 Coverage Extension . . . . . . . . . . . . . . . . . . . . . . 1045.5.3 Routing performance . . . . . . . . . . . . . . . . . . . . . 106

5.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 108

6 Conclusion and Future Directions 1096.0.1 Future directions . . . . . . . . . . . . . . . . . . . . . . . 111

A Resume du manuscrit de these en francais 115A.1 Vers un protocole de routage geographique avec contention et com-

munications cooperatives pour les reseaux de capteurs . . . . . . . 115

B La Problematique 117B.1 Contexte et defis . . . . . . . . . . . . . . . . . . . . . . . . . . . 118B.2 Contributions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

C Contributions 121C.1 CoopGeo : A Cooperative Geographic Routing Protocol . . . . . . 121

C.1.1 Modele de communication . . . . . . . . . . . . . . . . . . 121C.1.2 CoopGeo : A geographic cross-layer protocol for coopera-

tive wireless networks . . . . . . . . . . . . . . . . . . . . . 124

16 Contents

C.1.3 Transmission greedy sans balises de controle (BLGF) . . . 125C.1.3.1 Selection a base de temporisateurs (TCBF ) . . . . 125C.1.3.2 Recuperation greedy sans balises de controle (BLRF)126

C.1.4 Selection du relais base sur des informations geographiques 126C.2 Evaluation des Performances . . . . . . . . . . . . . . . . . . . . . 127

C.2.1 Taux d’erreur de paquets (PER) . . . . . . . . . . . . . . . 128C.2.2 Probabilite d’erreur dans la transmission de bout en bout . 129C.2.3 Variation des parametres d’entree . . . . . . . . . . . . . . 129

C.3 RACR : Relay-Aware Cooperative Routing . . . . . . . . . . . . . 131C.4 SER-Based Radio Coverage Formulation . . . . . . . . . . . . . . 131C.5 Architecture de RACR . . . . . . . . . . . . . . . . . . . . . . . . 134

C.5.1 Selection du relais . . . . . . . . . . . . . . . . . . . . . . . 136C.5.2 Selection du noeud intermediaire . . . . . . . . . . . . . . 136

C.6 Evaluation des performances . . . . . . . . . . . . . . . . . . . . . 137C.6.1 Extension de la couverture . . . . . . . . . . . . . . . . . . 137C.6.2 Efficacite energetique . . . . . . . . . . . . . . . . . . . . . 138

Bibliography 139

Glossary 151

List of Figures 153

List of Tables 157

Chapter 1Introduction

Wireless technologies has grown rapidly in the last decades and the advancesin hardware components have followed the same trend enabling the massive pro-duction of communication devices like laptops, cellular phones, personal digitalassistants (PDA), sensors, processors, etc. Since these devices are getting smallerand cheaper, their association with some technologies leads to the development ofnew kinds of wireless networks or the enhancement of existing ones such as cellu-lar networks (2G, 2.5G, 3G). Other type of wireless networks are the traditionalwireless networks that are spanned into infrastructure and infrastructure-less likead-hoc, sensor and mesh networks. These new network types, support a certainnumber of applications, including WLAN hotspots, real-time communications,home networking, surveillance systems, industrial control, vehicular networks,sensor networks and many other. Data wireless network (making reference toIEEE 802.11 and derivations) have been the center of research and commercialinterest for several years, nowadays, we can find Wi-Fi services almost every-where.

In spite of the industry and users interest oriented to traditional wirelessnetworks, recently the scientific and industrial community is turning into a dif-ferent scenario where a spontaneous group of electronic devices try to communi-cate without any infrastructure: the ad-hoc and sensor networks. They presentimportant challenges in their architecture design that are inherited from theirbasic characteristics. Ad-hoc and sensor networks are wireless, self-organizingsystems formed by nodes in communication range of each other forming a tem-poral network. In general, ad hoc networks are formed dynamically by static ormobile nodes that are connected via wireless links excluding a centralized admin-istration or a network infrastructure, the nodes can joint and quit the networkspontaneously, thus a method to organize themselves is needed to support all

17

18 Chapter 1. Introduction

the topology changes providing without interruption the routing paths from thesource towards the destination, briefly, the routing philosophy is to cooperatebetween nodes to forward the data in a multihop manner.

Wireless sensor networks (WSN) belong to the class of ad hoc networks, butthey have some extra characteristics that make them a special case, even if sen-sor networks closely looks like the behavior of ad hoc networks, sharing manychallenges as limited energy available to each node and the error-prone channels,they have differences to keep in mind: for example, the small size of nodes ina WSN involve that the nodes have very limited resources such as, processingspeed, memory, energy, and transmission power; as the nodes can be small, theycan be deployed in very large quantities, presenting higher node densities. We canalso add that they are inexpensive, the nodes are unattended after deploymentand designed for a prolonged lifetime with no maintenance or troubleshooting,and shorter transmission ranges.

Thus, WSN are rather designed to detect events or natural phenomena, nodescollect the data, process them and transmit to some users. By the nature of thosecharacteristics, the most challenging issues in sensor networks are the efficientuse of limited resources and the adaptable to topology changes derived from thecondition of the communication channel.

1.1 Context and challenges

The present work is substantially oriented to the wireless sensor networks,even if some protocols, algorithms and techniques are also applied to other typesof wireless networks. Thus, in order to clearly define the scope of the work wepresent the basic characteristics and challenges attached to our subject understudy.

The wireless Sensor Networks, are composed of a large number of sensor whichare small in size and are able to sense, process and communicate with each otherwhere the main finality is to detect events or phenomena, collect and processdata and transmit the information to the users. In addition to the characteristicsrelated to the wireless communication, the sensor networks include some otherbasic characteristics inherited from the way they work:

• Self-organizing capabilities• Short range communication and multihop routing• Dense deployment and cooperative effort of nodes• Frequent changing topology due to fading, mobility and node failures• Limitations in energy, transmit power, memory, and computing power• Data centric communication model

1.1. Context and challenges 19

In the WSN applications classes, we mention some of them to accentuate thedifference with other kind of wireless networks, such as:

• Infrastructure security• Environment and habitat monitoring• Industrial and agriculture sensing• Traffic control• Disaster prevention• Medical care

From the above summary, we can remark the data centric communicationmodel that influence the design of routing protocol, where all the data generatedby the nodes is forwarded to some data collection points commonly known assinks. With this communication model in mind, we can say that the designof wireless sensor networks has some important tasks to fulfill such as, controlthe consumption of limited resources that are affected by the transmissions andreception of data packets, and to limit the wireless channel effects generated bytwo physical phenomena: 1) From multipath propagation of the electromagneticwaves that generates variations in the received signal strength as function of thenode location and frequency. 2) From the influence of the eventual motion ofnodes that produce the effect called wireless channel variation, generally calledfading. Thus, a rigorous protocol design is imposed with the goal of minimizethe resources consumption (ie. energy) and maximize the network lifetime withthe constraint of detecting the topology changes so as to maintain the networkconnectivity and calculate the proper routes from the sensor nodes to the sinks.

In order to effectively tackle these WSNs challenges, the research communityhas produced a lot of works, focusing on traditional layered approaches, whereeach layer of the protocol stack (ie. network, Medium Access Control and physi-cal) is unaware of the operation of the other layer, eliminating thus, the benefitsof joint optimization across protocol layers which can improve the network per-formance.

Therefore, the use of a cross-layer design must be mandatory, where the avail-ability of some important information among the stack layers allows a node tomake more effective routing decisions as it will have a wider view of the networkbehavior, resulting in an improvement in the global network performance.

This dissertation, take the cross-layer approach as reference to enhance thelayered approach and propose an architecture designed to analyze the interactionbetween the network layer to route de packet to the destination, the mediumaccess control (MAC) layer to get access to the wireless channel and the physicallayer to adapt the protocol to the wireless conditions of the environment. Weshow in our contributions, that a cross-layer architecture may optimize the inter-


actions between these three layers and achieve a high performance and reliablecommunication in the network.

1.2 Contributions

During this research work, we consider the Geographic Routing as a concretesolution to the routing issue where the sensor nodes builds on a local manner theroute to the sink node. In this scope, the main goal of this thesis is to fill thegap between the traditional geographic routing protocols and the physical envi-ronment where the sensor nodes are located. To reach this goal, we approach therouting problem with lossy links by the use of a Beaconless Geographic Routingapproach to build the route not only in a local manner, but also on the fly andthe node cooperation approach as long as, it is an attractive scheme since it usesthe cooperative communications to exploit the broadcast nature of the wirelesschannel, allowing to the radios located at each node to jointly transmit informa-tion through relaying. Thus, our primary contributions are:

• A cross-layer design framework called CoopGeo (Cooperative Communica-tions for Geographic routing) is proposed. CoopGeo has been extensivelyevaluated and compared to a traditional beaconless routing protocol in sim-ulations. CoopGeo performs the greedy forwarding mechanism without us-ing beacon messages. Instead, each node broadcasts the message and eachreceiving node competes to forward it based on its local metric. Once theforward node wins the contention to transmit the message, we eventuallyapply a cooperative communication with single relay selection mechanismwhere the source node and relay node will jointly transmit the informa-tion in the wireless channel. With CoopGeo, we improve the physical layerperformance in terms of reliability, extending also the network lifetime com-pared to other geographic routing protocols which do not consider the MAClayer issues. We also apply a mechanism to get out from a local minimumproblem using a beaconless planarization mechanism.

• The RACR protocol exploits the coverage extension as a result from nodecooperation to improve the non-cooperative geographic routing. It is analternative to scenarios where network resources like energy should be pre-served while respecting a Symbol Error Rate constraint (SER). Thus, theproposed routing protocol, enables a node to make a local route decisiondepending on the geographic location of a relay while this relay is selectedwith the purpose of providing the maximum coverage extension toward the

1.3. Thesis Outline 21

destination.

The results obtained from the evaluation of CoopGeo and RACR makes usthink that both solutions are applicable to sensor networks in presence high mobil-ity or very variable channel environments. Thus, we can say that our cross-layervision of the problem may provide an integrated solution to problems like defineefficient routing paths, medium access, localization, or fault tolerance.

In addition to the main contributions, our research work provides the follow-ing complementary contributions:

• An extensive state of the art where we wanted to establish some impor-tants links between the geographic routing protocols (with its derivation:beaconless geographic routing and virtual coordinates) and the cooperativecommunications.

• We integrate two different beaconless geographic contributions into oneframework. We took the traditional beaconless greedy forwarding fromone contribution and the beaconless recovery forwarding from other contri-bution, and we formed one framework that we used to integrate over it outfirst contribution.

1.3 Thesis Outline

This thesis is organized as follows.

• Chapter 2, first, reviews the state of the art on routing in wireless sensornetworks, giving special interest to geographic routing and beaconless geo-graphic routing. Then, it presents the cooperative communications funda-mentals, addressing their advantages, their challenges, and their drawbacks.It lastly, gives an overview of the related works covering relay selection oncooperative communications for sensor networks.

• Chapters 3 covers the theory of the models discussed and used in thepresent work, such graph models, radio propagation models, mobility mod-els, etcetera. This information will help us to allow the appropriate under-standing of the contributions. We also present the tools used to analyze theperformances of the routing protocols.

• In chapter 4, we presents the details about our first contribution CoopGeo.This cooperative strategy builds a reliable Beaconless Geographic Routing


protocol for Wireless Sensor Networks using a contention-based relay selec-tion mechanism. we also introduce the model and the policies for selectionthe relay nodes according to a predefined metric.

• In chapter 5, we describe our second contribution, RACR, this is a routingprotocol that takes advantage of coverage extension offered by cooperativecommunications to achieve a scalable and efficient sensor network perfor-mance.

• Chapter 6 summarizes our contributions, discuss the perspectives and fur-ther improvement to the contributions and finally it concludes this thesis.

Chapter 2State of the art

In the first part of this chapter, we provide an introduction to routing insensor networks and we also supply an overview to the prior relevant works tothe subject developed in other parts of the thesis. We start by describing generallythe existing approaches to the routing problem. We then cover the GeographicRouting (GR), presenting the way how it works, the advantages, the challengesand the limitations. Then we present a Beaconless Geographic Routing (BLGR)as a derivation of the traditional GR, giving some specific references to otherrelated works.

In the second part, we introduce the cooperative communications as a mech-anism to help the wireless communications. We cover a brief overview and intro-duction to this communication scheme and present why they represent a lot ofinterest in the design of communication protocol. We then cover the backgroundin several basic areas: communications theory, graph models, and radio propaga-tion models in order to provide the basic ideas used in this thesis and allow thusits appropriately understanding.

2.1 Routing Protocols for Wireless Sensor Net-

works

The basic tasks of sensors comprised in a WSN are sensing or gathering theinformation from the analyzed phenomenon, processing, and transmitting thesensed data. To deliver data to the sink, sensor nodes use the multihop principlewhenever infrastructure is unavailable or as the direct transmission from thesource to the destination is almost impossible, thus the node sensing the datause intermediate nodes to relay its packets across the network until it reaches

23

24 Chapter 2. State of the art

the destination. Therefore, the source and the intermediate nodes have to decidewhich neighbor the packet will be sent, using a routing protocol as a tool to makethis decision.

Routing is still a challenging issue in mobile ad hoc networks and particularlyin sensor networks [Akyildiz 02], due to their limited resources, the nodes mobilityand the physical environmental conditions. In these kinds of networks, the mosttraditional routing protocols are classified into two categories: the topology-basedand position-based routing protocols.

2.1.1 Topology-based routing

Topology-based routing protocols use the information about the neighborsor links existing in the network to perform packet forwarding. They can befurther divided into proactive, reactive, and hybrid approaches. Proactive algo-rithms employ classical routing strategies derived from wired routing protocolssuch as distance vector or link state protocols. To illustrate the former strategy,we can cite from [Perkins 94] the Destination Sequenced Distance Vector proto-col(DSDV) and for the latter strategy, from [RFC 03] the Optimized Link-StateRouting protocol(OLSR) or from [Ogier 04] the Topology-Based Reverse PathForwarding (TBRPF). The protocols belonging to this approach try to keep aconsistent and up to date routing information about the paths available in thenetwork by exchanging control messages at fixed time periods. The main draw-back of this approach is that the maintenance of routing tables may occupy asignificant part of the available bandwidth and nodes resources, and this is evenworse when the topology of the network changes frequently.

In contrast to this problem, reactive protocols establish a route to a givendestination only when a node requests it by initiating a route discovery processand once the route is established, the data communication starts, thereby theyreduce the resource consumption of the network. This tend to work well insmall networks, however, this scheme does not scale well in large networks dueto the significant overhead generated to find the paths between the source anddestination nodes and the delay to transmit the data packets. Even if routemaintenance is restricted to the routes in use, the process can generate overheadin the network when the network topology changes very frequently or the packetcan be lost in the transmission process if the new route is not known. Examplesof this kind of protocols are Dynamic Source Routing (DSR) [Johnson 04] andAd-hoc On-Demand Distance Vector (AODV) [Perkins 03].

Finally, hybrid approach applies principles of both routing schemes, proactivefor the local neighborhood and reactive for the distant nodes or they may modifytheir behavior given some circumstances, sometimes using proactive rules andother times using the reactive rules. These kind of protocols try to mix combine

2.2. Geographic Routing Protocols 25

the power of both schemes, some well-known examples are Zone Routing Protocol(ZPR) [Haas 97], Location-Aided Routing (LAR) [Ko 00] and Distance RoutingEffect Algorithm for Mobility (DREAM) [Basagni 98].

2.1.2 Position-based routing (Geographic Routing)

A different vision from topology-based protocols, is the one presented by theposition-based protocols that from now and on, we will call geographic-basedrouting protocols. They use the node location to route the information, takingadvantage of the presence of low power GPS receivers at sensor nodes or theuse of other techniques to build a coordinate system with virtual or absolutecoordinates. Geographic routing in essence works at the network layer providingthe way how a packet will be delivered to the destination or sink node basedonly on local information, for instance the node location. Hence, geographicrouting [Stojmenovic 02] is an attractive approach to route in ad hoc networksand wireless sensor networks, because it is efficient, scalable, and it does not needto know the topology of the network.

Unlike topology-based protocols, geographic-based protocols require very fewcontrol traffic to work and they do not need to create or maintain the wholeroute from source to destination and thus they eliminate the overhead of frequenttopology updates.

Many Geographic-based routing protocols have been proposed in the liter-ature. We first describe the geographic basic principles. Then, we state theirbenefits, their forwarding strategies. Next, we present some geographic routingprotocols, and finally, we discuss their drawbacks.

2.2 Geographic Routing Protocols

In this section, we present some principles used in the development of geo-graphic routing protocols for wireless sensor networks. Some of them, have beenrefined and improved, but usually, some basic ideas remain the same such as:• Each node knows its geographic location using some localization mechanism

or hardware composant. Location awareness is essential to this routingapproach, so it is expected that wireless nodes will be equipped with a GPSor will execute some localization techniques. Several techniques exist forlocation sensing based on proximity or triangulation using radio signals,acoustic signals, or infrared.• Each node knows its direct neighbors locations (1-hop neighbors). This in-

formation could be obtained by a periodic information exchange containingthe position in form of beacons, hellos or broadcast messages. It is impor-


tant to note that this assumption could be relaxed and give birth to anothercategory of geographic routing.• Each node trying to send data packets already knows the destination loca-

tion.

Moreover, we state several benefits associated to geographic routing protocols:• Low overhead, since the establishment and maintenance of routes are not

required in protocols that uses location information to route data.• Localized algorithm, since a node only interacts with other nodes in a re-

stricted vicinity to determine which neighbor will forward the message.• Scalability, since these protocols are localized, they neither need to know

global information nor recompute the routing tables to follow the topologychanges.• High performance since they can adapt rapidly to the network dynamic,

respecting resource constraints of the nodes.

In geographic routing, the source and intermediate nodes forward the packetswith respect to a predefined routing metric, in general: the geographic progressto the destination. This kind of protocols have a ”greedy” behavior, selectingthe neighbor that minimizes the distance to the destination. However this greedymechanism could fail when it gets into a local minimum: a packet may be stuckat a node that does not have a neighbor closer to the the destination than itselfto forward the packet.

Figure 2.1: Greedy Forwarding: Node s forwards the packet to neighbor F1

Traditional geographic routing protocols also need to send beacons messagesperiodically to get their direct neighbors position and execute the greedy mech-anism, however, they may present problems with the mobility of nodes and evenif they can adapt the beacon frequency to the degree of the network mobility inorder to keep the nodes position update, they can still suffer the inaccurate posi-tion problem. Therefore in the presence of high mobility, the inaccurate position


information can lead to a significant decrease in the packet delivery rate and fastenergy consumption in wireless nodes due to the media access control (MAC)layer retransmissions.

2.2.1 Forwarding Strategies ”Greedy Forwarding”

The basic principle of geographic routing is that, when a node wants to senda packet, it includes the position of the recipient in the packet. Thus, when anintermediate node receives a packet, it forwards the packet to a neighbor lyingat a location in direction of the recipient. Ideally, this process can be repeateduntil the recipient has been reached (see Fig. 2.1). The forwarding strategyused, is an important part of the geographic protocol in use, since it allows tochoose among the neighborhood the node that will forward the packet. Thus,we proceed to present the more common forwarding strategies used by somegeographic protocols.

Fig. 2.2 depicts a scenario where S transmit a packet towards the destinationnode D, the circle with radius r indicates the maximum transmission range of S.The geographic strategy proposed by Takagi and Kleinrock [Takagi 84] introducesthe notion of ”progress” where intuitively the node forwards the packet to thenode that is closest to D. In the scenario the chosen node is A where the progressis calculated by the distance between the current node S and the projection A′ ofthe neighbor A onto the line formed by S and D. This strategy is known as mostforward within R (MFR) and tries to minimize the number of hops a packet has totraverse in order to reach D. MFR is a good strategy in scenarios where the senderof a packet cannot adapt the signal strength of the transmission to the distancebetween sender and receiver. Fin [Finn 87] proposed the strategy that nowadaysis known as ”GREEDY” which minimizes the distance d to the destination D (e.g.node B in Fig. 2.2). This is the most used scheme in geographic protocols foundin literature. In nearest with forward progress (NFP), the packet is transmittedto the nearest neighbor of the sender which is closer to the destination. In Fig.2.2 this would be node A. If all nodes employ NFP, the probability of packetcollisions is reduced significantly. Therefore, the average progress of the packet,calculated as p∆f(a, b) where p is the likelihood of a successful transmissionwithout a collision and f(a, b) is the progress of the packet when successfullyforwarded from a to b, is higher for NFP than for MFR.

Kranakis [Singh 99] proposed the compass routing, which considers the anglebetween the next hop, current, and destination nodes. This means that, the nextforwarding node minimizes the deviation from the line connecting the current andthe destination node (e.g. node D in Fig. 2.2). From this angle-based scheme(see Fig. 2.3), we can also mention Nearest neighbor routing (NN), where given aparameter angle α, node S finds the nearest node V as forwarding node among all


Figure 2.2: Forwarding strategies

neighbors of S in a given topology such that ∠vsd∠α and the Farthest neighborrouting (FN), where given a parameter angle α, node S finds the farthest nodeV as forwarding node among all neighbors of S in a given topology such that∠vsd∠α.

Figure 2.3: Nearest and farthest neighbor strategies

Finally, Nelson and Kleinrock [Nelson 84] proposed a random progress methodwhere the sender randomly choose one of the nodes with progress and forwardsthe packet to that node. This strategy relax the accuracy of information neededabout the position of the neighbors.

We have described in general terms the operation of the geographic routingprotocols. In the next subsections, we describe several protocols based on theabove principles and for the sake of simplicity an taking into consideration theirspecific characteristics, we can classify them in Beacon-based geographic routing(the traditional protocol, commonly called geographic routing) and Beaconlessgeographic routing. We describe them in the following subsections.


2.2.2 Some Traditional Geographic routing protocols

Even if this work is orientated to the beaconless geographic routing, the basisof its way of work comes from the traditional geographic routing protocols, thus,we consider important to describe some works that marked some trends anddirections in the field.

2.2.2.1 Greedy Face Greedy (GFG)

GFG [Bose 01] is the first protocol that merges and enhances two basic prin-ciples presented in section 2.2.1. They use the greedy routing proposed by Finto send the packet to forward to the node closer to the destination, minimizingthe distance to arrive to the target node (remember that they only consider thegreedy forwarding since a backward node may lead to a loop in the packet de-livery), and the face routing from Bose et al to bypass the holes in the networktopology and avoid the loops as well (see Fig. 2.4). This is the first geographicalgorithm that guarantees the packed delivery to its destination. The main prin-ciple of GFG, is that when a node finds a local minimum, it routes the packetin the interior faces intersected by the straight line connecting the source nodeand the destination, the way how the packet traverses the faces is by applyingthe right or left hand rule. Thus, when a packet is forwarded along an edge inclockwise or counterclockwise direction from the edge where it arrives. In thecase of, a packet arrives at an edge that intersects the imaginary line between thesource and the destination, the next face intersected by this line is handled in thesame way. In brief, GFG starts working in greedy mode and switch to recoverymode applying the face routing when a hole in the topology is found.

2.2.2.2 Greedy Perimeter Stateless Routing (GPSR)

GPSR [Karp 00] is an algorithm that appeared after GFG, but it actually hasa very similar operation since the same algorithms are applied (see Fig. 2.4).The contributions of GPSR with respect to GFG are: the integration of greedyforwarding and face routing into the IEEE 802.11 using thus, a more realisticmedium access layer instead of the ideal one used by GFG, and the discussionabout the Relative Neighborhood Graph (RNG) as an option to the GabrielGraph (GG). Afterwards, the Embedded Networks Laboratory team at the Uni-versity of Southern California (USC) implemented GPSR to run TinyOS. How-ever, a small difference between GPSR and GFG is that GFG may stay in thesame face after an intersection in the face routing is encountered while GPSRalways changes to the next face.


Figure 2.4: In blue the right hand rule and in red the face changes, two principlescomposing the face traversal on a planar graph used in GFG and GPSR algorithmsstrategies

2.2.2.3 Other Geographic protocols derived from greedy + face union

Kunh at al, proposed three contributions that are closely related to the greedyforwarding and face routing union. First, in the Adaptive Face Routing AFR[Kuhn 02], Kunh basically enhances the face routing phase by the employmentof an ellipse whose size is iteratively incremented as needed and is bounded bythe complete optimal path from the source to the destination when applyingthe recovery mode. In contrast, the original face routing needs to explore thecomplete boundary of the faces. In AFR, if the exploration hits the ellipse,it has to ”turn back” and continue its exploration of the current face in theopposite direction until hitting the ellipse for the second time, which completesthis face’s exploration. Secondly, Kunh proposed Greedy Other Adaptive FaceRouting (GOAFR) and Greedy Other Adaptive Face Routing Plus (GOAFR+)in [Kuhn 03b, Kuhn 03a], where they use the greedy and a modified version oftheir AFR contribution. In GOAFR they use the percolation theory and describethe worst case optimal and average case efficient. GOAFR+ is an enhancementto GOAFR for the average case and they also analyzed some cost metrics. Thus,we can say that GOAFR+ use two mechanism: early fall back and boundarycircle. The former, is used to return from face routing mode to greedy forwardingmode as soon as possible. The latter is used to restrict a searching area in similarway as AFR achieving optimal results in the worst casa. Briefly, the main thedifference between GOAFR and GOAFR+ is that GOAFR does not include thefalling back mechanism presented by its enhancement version.

2.2.2.4 Greedy Distributed Spanning Tree Routing (GDSTR)

Leon et al in [Leong 06], proposed a different approach than traditional geo-graphic routing. They describe an algorithm that does not require the planariza-tion of the network in recovery mode by routing on a spanning tree until it reaches


a node where the greedy mode can be applied again. GDSTR uses the convexhulls to aggregate the locations covered by the spanning tree and decide efficientlythe direction of the tree where the node must forward to make a progress to thedestination. Briefly, GDSRT build a spanning tree and every node maintains aconvex hull, when a node can not forward any more in greedy mode, it checksif the destination is found in its convex hull. If this is the case, it forwards thepacket to the proper child node, otherwise, it sends the packet to its parent thatmaintains a bigger convex hull. Thus, when a node finds another node nearer tothe destination than the node where the recovery mode started, it changes to thegreedy mode.

2.2.2.5 Local Tree based Greedy Routing (LTGR)

In [Liu 07b], Liu et al follow the approach proposed in GDSTR in order toavoid the face routing that uses a planarization method to route the packets. InLTGR, when a node gets into a local minima, it uses a local tree instead of facerouting without making face routing assumptions such that, the uniform radioranges of nodes and the bi-directional links existence that hard to be found in realworld networks. When a node sends a packet to a destination and this packet isblocked at a local minima, a local tree that can expand to a spanning tree coveringall nodes in the network is created to find the next hop toward the destination.The challenges found by the authors is how to embed local tree information inthe packet that is used by the data packet when traversing the local tree. Theyproposed two solutions: 1) When a node receives a packet marked in recoverymode, it divides the space in four quadrants. Then it adds one neighbor byquadrant (except the quadrant that contains the node where the packet comesfrom) to the local tree. Thus, a node can have only three children in the tree,controlling the number of nodes participating in the routing problem. If the nexthop node is not found in the local tree, it can be extended with another ”wave”of children nodes. 2) They propose a compression technique to embed the localtree information stored at the data packet.

2.2.3 Beaconless geographic routing (BLGR)

Traditional geographic protocols need periodic exchanges of hello messages inorder to aware neighboring nodes about its current position information. Thisis a proactive mechanism that leads to energy consumption in nodes, and thisconsumption could be increased if the hello interval is reduced to maintain theinformation updated due to a high mobility of the nodes.

Accordingly, Beaconless Geographic Routing (BLGR) is an enhancement tothe traditional geographic routing approach that overcomes the problems pre-


sented by the latter in high mobility scenarios. This approach is based in a prin-ciple where the node detaining the packet to transmit is not aware of the existingneighbors it has. So, it neither needs to have a previous knowledge of the networktopology to forward the data packet avoiding therefore the exchange of controlmessages nor information about the available routes to the destination. Beacon-less routing uses an scheme where the decision to forward is delegated to theneighbor nodes using only local information. In such scheme, all the candidatesto forward the message participate in a contention process where all neighborsstart a timer with respect to the progress to the destination and thus, the nodewhose timer expires the first will forward the packet to the destination.

We can say that the basics of the routing scheme can described as follows:• A node s that wants to transmit a message to d, choose a forwarding area

where it tries to find a potential forwarding node.• s broadcast the packet or the control message to start the packet transmis-

sion.• Nodes in the forwarding area, hear the packet and start up a timer in order

to become the forwarding node if the destination is not reached. In general,nodes that are not located in the forwarding area discard the packet (thisstep may change in accordance to the type of beaconless algorithm used.• The node whose timer expires the first will be the forwarding node and the

process start again the step 1 until the destination is reached.

The difference between the algorithms belonging to this approach are: howthey choose the forwarding area, how they deal with empty areas, how avoidthe collision between the node that contents for the transmission, and the metricused in nodes located at the forwarding area that defines the best suited node totransmit the packet.

As traditional geographic protocols, the greedy forwarding and the recoverymodes are the main modules of beaconless routing. In the former, the decision isgenerally based on the progress to destination (minimizing the hop count). Forthe latter module that is used to avoid the local minimum problem, two differ-ent approaches are used: heuristic, where the procedure does not theoreticallyguarantee the packet delivery and neighbor planarization where a face routingalgorithm is used and the message delivery is theoretically guaranteed.

A common behavior in general BLGR schemes is that, as they use the progressto destination as the principal metric when the nodes take the forwarding decision,obtaining this information at the network layer or at the Medium Access Controllayer but they are not attentive to the physical conditions of the environment.

The first works in Beaconless Geographic Routing were realized by Heis-senbuttel and et al in [Heissenbuttel 03], who proposed the Beaconless Routing(BLR), Fussel et al. in [Fubler 03] proposed Contention-Based Forwarding (CBF)


and in [Blum 03] Blum et al. presented the Implicit Geographic Routing (IGF).All these protocols are very similar, they all apply a mechanism to select the nexthop node in a distributed manner without knowledge of its neighborhood whichis based on contention timers. The difference are, for CBF and IGF, they focuson the integration of the routing protocol to the MAC layer, in this case IEEE802.11 was used. In BLR, the authors analyzed the theoretic properties inher-ent to this kind of protocols. They also propose, several enhancements to theoriginal design such as: the convergence from broadcast to unicast packets, theaggregation of paths and two recovery modes in case of local minimum problem.

2.2.3.1 Beaconless Routing (BLR)

In BLR [Heissenbuttel 03], the authors defined a forwarding area restrictedto a 60 sector from the node detaining the packet to the destination with aradius that just equals the transmission range r. They establish the 60 angle asa precondition since they consider that each node within this sector should beable to detect the transmission or any other node within this sector and thereforethe algorithm would work properly. In Fig. 2.5 we depict an scenario whereonly nodes within the sector take part in the competition process to forward thepacket (e.g. nodes A, B and C). The rest of the nodes do not participate in theprocess and just discard the packet (e.g. nodes D, E and F )

Figure 2.5: 60 sector from S to D within the transmission range

To forward the packets, BLR define a Dynamic Forwarding Delay (DFD) ateach node that is related to its progress to destination, so to calculate the DFDvalue each node within the previously defined sector determines its progress pand infer ContentionDelay in the interval [0, ..,MaxDelay] which indicates thedelay introduced before trying to forward the packet.

Additionally, the authors proposed some enhancements to improve the al-gorithm performance. They converge the transmission once the node detaining


the packet already knows the next hop position through a previous transmissionfrom broadcasting to unicasting the packet during a predefined threshold. Theyalso proposed two recovery strategies for the local minimum problem (no node islocated at the forwarding sector), actually they call them backup mode. First, au-thors detail the request-response approach where the node in the local minimumsend a request packet and all neighbors answer indicating their location, then thenode open the forwarding area to an approximately 180 sector and wait for aforwarding candidate, if there is no node with forward progress, the actual nodeadopte build a planar subgraph and applies the right-hand rule (see descriptionin GFG, GPSR) via a unicast packet. Second, the clockwise-relaying approachwhere the node in the backup mode start again a contention timer according tothe angel α between itself, the previous node and the destination. Thus any nodewith forward progress transmit the packet before any other node with backwardprogress in a clockwise order.

2.2.3.2 Contention-Based Forwarding (CBF)

Fussler et al in [Fubler 03], deal with the traditional geographic protocols withtheir Contention-Based Forwarding scheme. In addition to the distributed nexthop selection using biased timers like BLR and IGF, they propose three suppres-sion strategies to handle the packet duplication or collisions between potentialforwarding nodes. CBF consists of two process: 1) the selection of the next hopby means of contention and 2) the suppression used to reduce the chance ofaccidentally selecting more than one node as the next hop.

CBF starts when a node s wants to transmit a packet to d, the node broad-cast its packet to its direct neighborhood, then the selection process begins bysetting out at each node timers with random values using uniform or exponen-tial distributions, then the candidate whose timer expires the first forwards thepacket. Consequently, they adopt the progress to destination to derive the timervalues instead of the random generation. Other candidate nodes cancel theirtimers when they listen to the winning node transmission toward d. During thisprocess, the define a minimum time interval needed to suppress other candidatesto forward δ and state that a packet duplication occurs when the best candidatenode has a progress P1 and there is at least another node with progress P suchthat t(P )− t(P1) < δ.

The suppression process tries to tackle the packet duplication and controlthe overhead (see scenario depicted in Fig. 2.6), Fussler et al propose threesuppression strategies as follows:• Basic scheme. This is the simplest and less performance than the other

two schemes. Here, if a node gain the forwarding contention, it forwards


the packet. When the direct neighborhood receives the message and thenodes belonging to this neighborhood that still have a timer running forthe packet, the timer is immediately canceled.• Area-based scheme. They propose to artificially reduce the forwarding area.

The key idea is to define the suppression area in such a way that all nodeswithin that area are in transmission range of each other, avoiding thus extrapacket duplications or collisions as they may appear in the basic suppressionscheme. In this direction, the author propose three areas: 1) the basic areadefined before, 2) a circle in direction of the destination with the diameterequivalent to the transmission range of the current node. 3) the reuleauxtriangle is a shape that better covers the forwarding area and thus offersmore forwarding candidate nodes.• Active Selection scheme. It is inspired by the Request To Send, Clear To

Send (RTS/CTS) mechanism from the IEEE 802.11 standard. Here, theforwarding node broadcasts a control packet called RTF (Request To For-ward) instead of immediately broadcasting the packet. The RTF containsthe forwarding nodes location and the final destinations location. Eachneighbor verify if it is located in the forward progress area, if this is thecase, it sets a reply timer according to the progress toward d. Thus, whena timer expiry occurs, a control-packet called CTF (Clear To Forward) istransmitted to the forwarding node and the nodes that hear the CTF sup-press their timers and the current node send finally the packet in unicastto the node which sent the CTF message.

Figure 2.6: Area-based suppression strategies. Node A,B and C are in thereuleaux triangle whereas only node C is in the circle area

2.2.3.3 Implicit Geographic Forwarding (IGF)

Blum et al in [Blum 03], proposed the implicit geographic forwarding wherethey integrate the routing and MAC layer into a common protocol, taking as basis


the IEEE 802.11 DCF contention scheme. The principle is the same as BLR andCBF; IGF tries to non-deterministically route packets using a competition processbetween candidate nodes. Compared to the two beaconless protocols explainedbefore, here IGF incorporate an energy remaining (ER) metric into the routeselection process. The procedure used to send a packet is the same as the activeselection described in CBF (using RTS/CTF control messages, but here theyare called ORTS/CTS -Open RTS-). The IGF forwarding area (see Fig. 2.5) isthe same as the one of BLR (a 60 cone towards the destination). In contrastto CBF, IGF provides a recovery process to bypass the local minimum problemimplementing a forwarding area shift where the network layer (previous MAClayer notification) retransmit a message to request a shift in the forwarding areato look for a next hop node and then applies the backpressure from [speed ref].In Fig. 2.7 we can see the sequence of shifts.

Figure 2.7: Forwarding area sequence shifts in IGF

2.2.3.4 Geographic Random Forwarding (GeRaF)

GeRaF [Zorzi 03], is another type of BLGR that takes the positive advance-ment area of the node coverage area into N sub-regions (see Fig. 2.7). Each ofthem delimited by the coverage area the the transmitting node and one or twocircle areas centered at the sink. Upon forwarding a packet, the sender broad-casts an RTS packet to all nodes in the sub-region closest to the destination andexpects a CTS reply. If there is no reply, the sender will broadcast another RTSpacket to the sub-region that is the second closest to the destination. This processcontinues until a CTS is received or all sub-regions have been searched in whichcase the forwarding fails. When nodes in the same sub-region answer to the RTSpacket, a collision may be produced, in this case, the authors adopted the useof busy tones as collision avoidance scheme. The collision avoidance scheme, the


nodes should be equipped with two radios to separate one channel for the datatraffic and the other is used as a wakeup signaling.

Figure 2.8: Regions from the destination point of view

2.2.4 New Beaconless Geographic Routing Protocols

In recent times, other extensions to BLGR were proposed, we present two ofthem that we consider they contribute with new ideas to the routing issue. Bea-conless On demand Strategy (BOSS) [Sanchez 07], extend the positive advance-ment subdivisions as GeRaf propose to the negative advancement area (see Fig.2.9) and thus, the shift from ”greedy” mode to recovery mode is automatically.In addition, they studied the impact of the packet size on the packet receptionrate by means of some practical experimentations. Kalosha et al [Kalosha 08],proposed two algorithms appropriate to the beaconless recovery process. TheBeaconlees Forwarder planarization (BFP) and the angular relaying, these con-tributions planarize the neighborhood to allow the face routing following thebeaconless principle to minimize the overhead of control messages.

2.2.4.1 Beaconless On Demand Strategy for Geographic Routing inWireless Sensor Networks (BOSS)

As stated in the above section, BOSS [Sanchez 07] considered in the protocoldesign the realistic physical layer and defined a delay function in order to mitigatethe collisions and duplicate packets from the next hop forwarding selection. Basedon some experimentations, they found as others before that the packet size has adirect relationship with the probability of error. As big packets have less proba-bility of being received than small packets, the authors decided to avoid the useof RTS/CTS control packets to start the communication between the forwardingnode i toward i+1 node since control packets may induce to select a node that infact, it may not receive the bigger data packet. To transmit a packet, BOSS usesa three way handshake. The node detaining the packet to send broadcast it andwait for the first response from the neighborhood and then, they confirms the


selection with a final control packet. To reduce the overhead, they use a passiveacknowledgement scheme to confirm the successful reception of the transmittedpacket. In case of nodes absence in the positive forwarding area, transparentlythe nodes located a the negative forwarding area form a planar subgraph wherethe face routing algorithm is applied to avoid the holes in the network.

Figure 2.9: BOSS extends the positive sub-regions from GeRaf to the negativearea of the node coverage area

2.2.4.2 Select and Protest-based Beaconless Georouting with guaran-teed delivery

Kalosha et al [Kalosha 08], based on thee select and protest principle, theyproposed two methods in the recovery mode to build a planar subgraph thatallows the direct application of the face routing algorithm and thus guaranteethe delivery of the packet to the destination. In the first contribution calledBeaconless Forwarder Planarization (BFP), they build an approximation of aplanar subgraph, then they sort out the false nodes from that approximation toget a final planar subgraph used in the beaconless recovery mode. In the secondcontribution called, the angular relaying, they try to directly find the next hopnode of the right-hand face traversal algorithm where if the selected neighbor doesnot belong to the planar subgraph, in this case the Gabriel Graph, the mechanismlook for a consecutive neighbor in the right-hand direction. Given this facts, wecan say that the authors proposed two mechanisms to route in recovery modebuilding planar or partially planar subgraphs minimizing the control messagesexchange between the neighborhood and the node detaining the packet to forward.

2.3. Virtual Coordinates 39

2.3 Virtual Coordinates

In previous sections, we have presented the geographic routing as one of themost efficient and scalable routing solution since it is stateless and localized: itdoes not need to recognize the topology of the network. To achieve this, each nodein the network must know its own coordinates and the destination coordinates.Such approach presents some drawbacks like the localization error that generatesrouting problems that could increase the path length or minimize the packetdelivery rate. Many sensor nodes do not have a Global Positioning System (GPS)embedded and providing such device to each node could be expensive in terms ofenergetic resources and money, and finally the network can not be deployed forindoor applications. Another kind of study was then derived taking the greedyforwarding principle of geographic routing protocols, but the greedy routing isexecuted by encoding the connectivity without geographical information, tryingto model the network as a geometric space and characterize the position of a nodeby a position in this space. Thus, the main idea is to build a virtual coordinatesscheme that replace the real physical coordinates. Hence, each node with thesecoordinates uses a predefined function to calculate the distance between itselfand its neighbors. Based on these distances, the greedy forwarding is appliedsimilarly to the traditional geographic routing protocols.

In this section, we present some important virtual coordinates protocols. Sincecommonly the virtual coordinates system construction is based on few nodes ofthe network (often called anchors or landmarks) that initialize the process. Wepresent a very fist classification based on their behavior. Hence, the protocolsmay to one of two categories: the first when the landmarks are location-aware andthe second where the landmarks do not know its coordinates (they are location-unaware).

2.3.1 Location-aware landmarks

The protocols in this category use some landmark nodes who know their physi-cal coordinates, and from these landmarks, the rest of the nodes infer their coordi-nates. To do this, the ordinary nodes execute some physical measurements and/orsome localization protocols such as triangulation and multi-lateration methodswhich are helped by other techniques to determining distances between the nodes,including Received Signal Strength (RSS), Time of Arrival (ToA), Angle of Ar-rival (AoA). This kind of protocols are not considered in the present work sincethey are designed just to locate a node in a place and then, another protocol isapplied to route the information. Refer to [Niculescu 04], [Langendoen 05] and[Savarese 01] for more information.


2.3.2 Location-unaware landmarks

This kind of protocols do not try to approximate their physical coordinates,since the landmark nodes do not know their physical coordinates. They ratherrelax this constraint and build a virtual system that encodes some network con-nectivity information instead of physical proximity. Hence, once the system isbuilt, the greedy forwarding is used over the virtual coordinates to route thepackets.

In this category NoGeo [Rao 03] based on the graph embedding conceptspropose to designate two landmarks, then perimeter nodes are detected and pro-jected in a virtual circle to compute their own coordinates, and finally the nonlandmark nodes calculate their virtual coordinates executing an iterative rubberband algorithm based on a centroid transformation. NoGeo is relatively complexas it requires a large amount of messages exchanges and iterations to converge.In addition to these drawbacks, they assumed that the network is static once theperimeter nodes are detected which trends to problems when new nodes appear.

Graph EMbedding for sensor networks (GEM) [Newsome 04] proposes a vir-tual coordinate system with polar coordinates (given in radius and angle from acenter). They first create the virtual coordinates system by constructing a span-ning tree, then the radius of each node is defined (number of hops from a nodeand the root), next each node is assigned an angle range (to identify the nodewithin a level) which will be respectively used to assign an angle range to itssubtree. Afterward, each subtree calculates its center of mass and propagates itto the root of the original spanning tree, resulting in a labeled graph tree withvirtual coordinates. At this moment, the routing process can be effectuated eitherby routing up in the tree until a node has been found that is a common parentof both source and destination or using a more efficient scheme called VirtualPolar Coordinate Routing (VPCR). Instead of always routing up the tree, VPCRchecks to see if there is a neighboring node with similar radius that is angularlycloser to the destination than the current node. If so, that node is given prefer-ence; if not, the packet is routed up the tree. Once an ancestor of the destinationis reached, the packet is routed downward just like in the simple routing scheme.

Later on, several research works appears, such as Logical Coordinate SpaceRouting (LCR) [Cao 04], Beacon Vector Routing (BVR) [Fonseca 05], VirtualCoordinate Assignment Protocol VCap [Caruso 05], and HopID [Zhao 05b]. Theyall have very similar behaviors: 1) Select few landmark nodes and span a tree fromthem to every node in the network; 2) Each node constructs its virtual coordinatesusing a vector (see Fig. 2.10) that represent the hop distance to each landmark3) A distance metric is defined so as to apply it when forwarding the packets. Insummary we can say that, first each node derive its virtual coordinates, and thenapplies the forwarding rule. The small difference between these works can be

2.3. Virtual Coordinates 41

found in the number of landmark nodes used, LCR defines four landmarks, BVRbetween ten ant eighty, and Vcap tree. Other dissimilarity is the technique used toavoid the loops in the routing path. BVR is the only protocol implemented in realsensor nodes and also while the other mentioned protocol defined the Euclideandistance as function as forwarding criteria, BVR applies the semi-Manhattandistance. Liu et al, stated later that neither the Manhattan nor semi-Manhattandistance are necessary better measures than Euclidean distance.

Figure 2.10: The node P builds its virtual coordinates vector using the threelandmark nodes (yellow)

In [Liu 08] Liu et al described an anomaly of the previous protocols since theintegral nature of the coordinate vector (virtual coordinates are based on integernumber of hops to the anchors) produces a quantization noise in the estimate ofconnectivity and node location. They improved the performance by proposingan Aligned Virtual Coordinate System (AVCS). To achieve this, they align thenodes by computing the coordinates of a given node as a function of its owncoordinates and the coordinates of the neighboring nodes. This new processallows the discrimination between nodes located at the same level of distancebetween the conflicting nodes and the landmarks.

Leon in [Leong 07] described the Greedy Embedding Spring Coordinates (GSpring)protocol inspired from the physical spring system with repulsion forces model.The model used points to an incremental adjusting of the initial virtual coordi-nates. In GSpring, each link is modeled like a spring by assigning forces amongthe set of edges between the nodes using the Hooke’s law. The forces are appliedto the nodes, pulling them closer together or pushing them further apart. Thisis repeated iteratively until the system comes to an equilibrium state. At the be-ginning, GSpring detects some perimeter nodes that will act like landmarks andassign them some coordinates, afterward, the rest of the nodes (ordinary nodes)compute their initial virtual coordinates. Then, the ordinary nodes update its co-ordinates based on the virtual coordinates of its direct neighbors using the springmodel.


2.3.3 Landmark-free Virtual Coordinates

Recently, in [Watteyne 09], they proposed the Centroid virtual coordinates, anovel protocol that we will refer as CVC. This new routing paradigm forces usto consider to add a new category in the virtual coordinates protocols classifica-tion: the landmark-free virtual coordinates. Watteyne et al, took the centroidtransformation used by Rao et al but they eliminate the initialization phase (nolandmark nodes needed). Instead, the nodes choose a pair of floating positivenumbers as coordinates except the sink node that has a fixed virtual coordinate,known by all the nodes. The virtual coordinates are refined (by the centroidtransformation) only when the node has some data to transmit, thus avoiding alot of traffic overhead. In summary the protocol works as follows. 1) The trans-mitting node retrieves tis neighbors virtual coordinates by means of an adaptedneighbor discovery procedure working at the MAC layer. 2) The current nodeupdates its virtual coordinates with the value obtained from the centroid trans-formation, and 3) The current node verifies that there is no neighbor virtuallyclose to him than a threshold distance, if that is the case a function is used toupdate the location of that neighbor. To apply the complete routing process froma source to a destination, they coupled the virtual coordinates protocol with the3rules [Watteyne 07] routing protocol and the 1-hopMAC [Watteyne 06] MACprotocol.

2.4 Cooperative Communications

Our intuition for solving some fundamental issues of wireless communicationslike fading and interferences neglected in common geographic routing protocolsis to focus on increasing the spectrum efficiency of the wireless channel by usingcooperative communications techniques. The former, emerges from the wirelesschannel variation whose origin lies in the multi-path propagation of wireless sig-nals that induce changes in the received signal strength as function of the nodemobility, location and transmitting frequency. The latter, emerges between trans-mitters communicating with a common receiver or between different transmitter-receiver pairs. In our research, our goal is to fill the gap between the geographicrouting protocols working at the network and MAC layer, and the physical layer.Therefore, we target the use of cooperative communications as a solution to theproblem described above.

In this section we explore the cooperative communications approaches, whichmitigate some wireless issues. These approaches, shift us to a new point of viewwhere multi-path fading can be viewed as an opportunity to be exploited to im-prove the performance of wireless networks and specifically for our case, the adhoc and sensor networks. Thus, our goal is to capture the main elements of

2.4. Cooperative Communications 43

cooperative communications, their properties, their application areas, and thusidentify the best suited approach to exploit the spatial diversity in wireless net-works as well as its appliance to multihop routing.

2.4.1 General concept

The cooperative communications refers to a set of distributed wireless nodesthat interact to jointly transmit information in a network where, radio terminalsrelay signals for each other emulating an antenna array and exploiting the spatialdiversity offered by the fading channels, creating additional paths between thesource and destination using intermediate relay nodes. Thus, the destination candetect the transmitted information in a wireless channel variation environmentsince from a statistical point of view, the chance that all channel links involved inthe transmission go down is rare. The main motivations to exploit some forms ofcooperation are: to improve the reliability of communications in terms of SymbolError Rate (SER), Bit Error Rate (BER) for a given transmission rate and alsoto increase the network performance with respect to the transmission rate.

The most simple way to describe the cooperative communication is to reducethe communication to a scenario with three nodes (see Fig. 2.11). This scenariois derived from the commonly the relay channel [Cover 79], where a coopera-tive node named relay, overhears a direct transmission between a source and adestination. First, the source node transmit its data, the destination and therelay node receive the data. Second, the relay node retransmit the previouslyreceived data from the source to the destination. Thus, the destination receivesthe same data twice, via two independent channel with different characteristics.Finally, the different versions of the data at the destination are combined usinga combination technique (ie. Maximum Ratio Combining, MRC; Equal RatioCombining, ERC; Fixed Ratio Combining, FRC; etc.) so as to obtain a finalversion of the transmitted data. Later on, others works have been derived fromthe previous model making a generalization of the relay channel where multiplerelays are used during the communication process.

2.4.2 Relay behavior and protocols

The relay nodes acting in cooperative communications may have several be-haviors with respect to how they process and decode the received signals as wellas the combining methods used at the destination nodes. In other referencesthey are called protocols or signaling methods. We summarize this behaviors asfollows.

In the Amplify-and-Forward behavior described by Laneman in [Laneman 04],the relay node receives a noisy version of the source transmitted signal. The


Figure 2.11: In phase 1, the source broadcast its data and relay and destinationreceive it. In phase 2, the relay node retransmit the data received in phase 1. Inthese two phases, three different fading paths are used

relay amplify the signal within its power constraint, and then retransmit it to thedestination which combine the received signals using a combining technique (ie.MRC). The main drawbacks of this behavior are: first, when the relay amplifiesthe signal, it amplify the noise as well. second, the relay must use the gain whichdepends upon the fading coefficient between the source and the relay, so somemechanisms to get or estimate this information must be considered.

In the Decode-and-Forward behavior also described in [Laneman 04], the re-lay node first receive the signal, decodes it, and then re-encodes and transmitthe signal. The decoding process may take two different forms: the relay maydecode the original signal completely which requires some computing resourcesbut has the advantage that an error correcting code could be used at the relayor it may simply realize a symbol by symbol decoding and let to the destinationthe full decoding task. Due to its simplicity and advantages with respect to theAmplify-and-Forward behavior, the Decode-and-Forward behavior is most oftenpreferred in cooperative networks. In addition to the presented behaviors, Lane-man introduces two kinds of relay behavior: selection and incremental relaying.But we rather consider this like an cooperative mechanism that may present oneof the previous behaviors. Hence, we will explain them in the next subsection.

2.4.3 Cooperative communications classification

Now that we have described the relay nodes behavior, we present a broadclassification of cooperation strategies that will help us to identify the strategyto follow in order to get an efficient spectrum utilization in scenarios based on


in Geographic Routing. This classification is composed by three approaches,described as follows.

In the fixed relay approach, all the nodes listening to the channel simplyapply either the Amplify-and-Forward or Decode-and-Forward behavior to relaythe data to the destination. In the case of fixed Decode-and-Forward behavior,the network suffers a lost in performance when the relay cannot decode the sourcemessage successfully.

The second approach is the relay selection, where the relaying nodes are able tomeasure the channel quality using the fading coefficients between the source andthe relay. In this scheme, a threshold is defined and the transmission is adapted tothe situation. Thus, if the measured coefficients falls below a assigned threshold,the source continues the direct transmission process using another mechanismlike the repetition or more powerful codes, so as to, the destination node cansuccessfully decode the original data. Otherwise, the relay nodes forward what ithad already received from the source with an Amplify-and-Forward or Decode-and-Forward behavior.

The third approach called incremental relaying, employs a limited feedbackfrom the destination node that sends a one-bit acknowledgment (ACK) to therelay and the source if it has decoded the source message successfully. Otherwise,it sends a NACK to indicate failure of transmission. In this case the relay, ifit has been able to decode the source message, retransmits the message to thedestination by employing repetition coding. This approach, can be viewed as anextension of the incremental redundancy or the hybrid automatic-repeat-request(ARQ) protocols that work at the link layer.

Another classification could be mentioned if we take into consideration thelayer where the cooperation takes place. Actually, nodes cooperation may occurin all layers of the OSI model. The cooperative techniques of our interest arelocated at the physical and network layer. The physical layer cooperation tech-niques, commonly known as cooperative relaying, ranges from the amplify andforward [Laneman 04], decode and forward techniques [Lai 06], to the multiplex-ing of signals into orthogonal subchannel as CDMA [Sendonaris 03] or TDMA[Laneman 04]. The network layer cooperation techniques, actually all multihoprouting protocols could be considered a cooperation techniques as nodes sharestheir resources to route some packet to the destination, optimizing a specific net-work metric (ie. path diversity, energy consumption)[Khandani 07, Ibrahim 08a,Zorzi 03, Biswas 04, Biswas 05].

An alternative vision of cooperation between nodes is expressed by the codedcooperation which is commonly known as network coding. Network coding presentsthe same motivations as the cooperative communications and relaying, but whilecooperative communications process signals, network coding process messagesor encoded packets, both visions are based on the transmissions over multiple


nodes but network coding area is out of the scope of this research work. Refer to[Ahlswede 00], [Li 03], and [Koetter 02] if you are interested.

From this very basic classification, a variety of schemes, techniques, and pro-tocols can be found. Some of them, are derived from the basic approaches, otherare combinaisons of the last two approaches where the relay nodes are establishedby a selection scheme that uses some feedback from the destination.

2.4.4 Relay selection

The selection and incremental relaying approaches have gained interest amongthe ad hoc and sensor networks research community because they can define anumber of relay nodes that will help the source to transmit the data. In general,the criteria used by the relays is: to forward the decoded information only if theamplitude of the measured channel coefficient of source to relay link is larger thana certain threshold. However, the challenge is how to define such a threshold thatallows the relay to forward only correctly decoded information.

In [Laneman 03], a Distributed Space-Time Coded (DSTC)cooperative proto-col is presented. In DSTC, all relay nodes that can decode the original transmis-sion re-transmit in the same subchannel using a designed space-time code. Thepotential relay nodes just have to estimate the Signal to Noise Ratio (SNR) ofthe received signal, and decode them if the SNR is above a threshold, re-encodethe signal and transmit it to the destination. Also in [Su 08], the author dis-cusses an alternate criteria that instead of setting a fixed threshold, the relay isassumed to be able to detect whether or not the information symbol is decodedcorrectly. Relay forwards decoded information to the destination if it can decodethe information symbols correctly. Otherwise relay stays silent.

Zhao in [Zhao 07], presented a group of several relay nodes which assist simul-taneously the transmission providing an additional Signal to Noise gain. Here, thenumber of relays represents the system diversity order. For instance, a schemewith m relays achieves a diversity order m + 1. Even if, the use of multiplerelays provide to the system a higher diversity order, it adds some complex-ity to the system that leads to a waste of bandwidth while adding difficulty tosynchronize the nodes. We can say that the cooperative relaying presented by[Laneman 03, Su 08, Zhao 07] where m relays operate, consumes node resourceslike energy and time that could affect seriously a network with constrained re-source like ad hoc and sensor networks. Thus, another approach taking intoaccount these constraints is needed.

Many other cooperative techniques involving multiple relays can be found inthe literature, for a broader vision, see [Sendonaris 03, Ibrahim 08b, Stanojev 06]


2.4.5 Single relay selection

The above issues, form a critical part in the design of relay selection protocolsfor wireless sensor networks, motivating thus the research community to continuethe studies about relay selection schemes but restricting the number of cooper-ating nodes in order to minimize the nodes resources consumption. Taking as astarting point the principle that only one relay is needed to cooperate with thesource node to transmit its messages, the design of the physical layer is simplified.Compared to [Laneman 03], the requirement of space-time codes is eliminated ifthe source and relay transmit in orthogonal time-slots. Hence, the above draw-backs can be minimized significatively. The present issues related to single relayselection is that it should tackle the problem of how and when to select a relaynode from a set of candidates and weather or not a cooperative communicationshould help the direct transmission.

In cooperative relaying, one new phase is added to the two traditional phasesof cooperative communication. It can be executed before the beginning of thedata transmission or between the direct and cooperative phases. This choiceis intrinsically related to the protocol design. The new phase is called relayselection, it is often based on the Channel State Information (CSI) between theparticipating nodes, and, usually the CSI is obtained for each transmission sincethe channel state may vary frequently.

Bletsas et al in [Bletsas 06], presented a single relay selection scheme basedon instantaneous local measurements. The idea is to let the node with the bestchannel condition relay the signal. Since only one relay is working at each timeslot, a very strict time and carrier synchronization among the relays is no longerneeded. Furthermore, because the transmission of one information bearing sym-bol is completed within two time slots, the relay selection has higher bandwidthefficiency. To select the best relay in end to end path between the source and thedestination, the potential relay nodes monitor the instantaneous channel condi-tions from themselves to the source and destination through the overhearing ofa RTS/CTS transmission. With the channel information estimation, the relayparticipate in a distributed relay selection procedure based on a timer. When arelay receives a CTS packet, it triggers a timer inversely proportional to the chan-nel estimation, so, the timer of the relay with best channel conditions will expirethe first. Then the self-selected relay sends a flag packet to signal its presenceto avoid other candidates try to relay the packet. Since a possible relay add twopaths (source-relay and relay-destination) to the transmission, and each of themhas different estimations, the authors proposed two policies to help each node toquantify its goodness to relay. In the first policy, the minimum estimation of thetwo hops is selected. In the second, the harmonic mean of the two paths is usedto build the timer. Given this facts, we can say that this protocol has a proactive


relay selection behavior since at the beginning the relay node is selected, thenthe source transmit its data, and finally, the relay re-transmit the data.

Zhao et al [Zhao 05a] added a diversity effect to the Geographic random For-warding protocol (GeRaF) [Zorzi 03] by means of a generalization of the Hybrid-Automatic Repeat Request (Hybrid-ARQ) and with incremental redundancy,which is served by the relay closest to the destination among those nodes thathave successfully decoded the message. The principle is that, retransmitted pack-ets do not need to come from the original source but could be sent by relays thatoverhear the transmission. In the protocol, known as HARBINGER -see also[Valenti 04] (Hybrid ARQ-Based INtercluster GEographic Relaying), a sourceencodes a b bit message into a codeword of n symbols length. The codeword isbroken into M blocks, the time is also divided in s = M time slots so that oneblock is transmitted at one time slot. At the beginning, the source transmit itsfirst block M = 1, then in the next slots s, s > 2, any node in the decodingset D(s) (nodes that have successfully decoded the message v ∈ D(s)), couldre-encode it and transmit the next block of the codeword. To choose the relaynode among those nodes in D(s), HARBINGER applies the same principle asGeRaF where nodes use the position information to identify the relays. Thus, itchooses the closest node to the destination by means of a distributed contentionphase.

On the same line of work, [Adam 08, Adam 09] attack the spectral ineffi-ciency and the channel over reservation in cooperative communications. Theypresented two adaptive relay selection protocols. In their first contribution, theyintroduced the Relay Selection on Demand with Early Retreat (RSoDER), whichuses a on demand scheme where the relay selection process is started only ifthey are needed by the destination node and an early retreat scheme where thenodes with bad channel conditions do not participate in the selection process.In their second contribution, they proposed Multi-Hop-Aware Cooperative Re-laying (MHA-Coop-Relaying) cross-layer scheme which tries to exploit routinginformation along with cooperative communications. Here, instead of using ahob by hop cooperation, they detected relay nodes that can act in a multi-hopenvironment (node relaying in a two-hops progress) to avoid the channel reser-vation. To achieve this, the source node should know the next two hops (ie. D1and D2) in the path to the final destination. Then, they put the potential relaynodes into two sets, the single-hop relay (potential relay nodes to D1) and themulti-hop relay composed of nodes that can relay the data in a two-hop distancefrom the source node (ie. potential relay nodes that are in D1 and D2 trans-mission range). Thus, when a message from source node gets D1 a traditionalcooperative communication has been effectuated. But, whenever the messagebeing routed reaches D2 during a cooperation process initiated by the source, atwo-hop cooperation is achieved improving the channel utilization.

2.5. Discussion 49

The Adam et al contributions are very similar to the work of Coronel [Coronel 07]presented one year before of the MHA-Coop-relaying appearance. The Coronel’swork differs in that, the protocol is based on geographic routing. Another impor-tant feature to remark, is that in MHA-Coop-Relaying the source node shouldknow its two next hops in the path to work, while in Coronel’s this informationis not needed.

In [Liu 07a], Liu et at described their Cooperative MAC protocol (CoopMAC)based on the IEEE 802.11 protocol. They presented two different versions: Coop-MACI and CoopMACII. The former, a new control frame Helper ready To Send(HTS) is added to prevent other nodes that a relay node will cooperate with thetransmission and the destination node reserve two channel slot times for bothtransmissions. In the latter, the HTS frame is not used, instead the RTS frameis used to advice to its neighbors who is the relay node.

Mainaud et al proposed WSC-MAC [Mainaud 08] for WSN. In their protocol,the nodes are randomly grouped into sets. When a node transmits a packet,the nodes with the same set ID will participate at the relay selection, thus theylimit the number of nodes involved in the process. Then a link state algorithmis executed at each node from the set that is assumed close to one, and the relaynode will cooperate only if it enhances the transmission performance with respectto the channel link estimation between the source-destination.

Other references to relay selection algorithms will be cited later while describ-ing our contributions.

2.5 Discussion

In this chapter, we have presented an extended survey of the state of the artabout the routing problem for wireless networks, specifically the sensor networks.Recent research works look at the geographic routing as an attractive solutionsince it has good scalability. However, most of the geographic protocols presentedhere suffers from some restrictions. They are implemented having in mind anisolated network layer and cannot cope efficiently neither with node and linkfailures or mobile nodes, nor with the wireless channel impairments like fadingand interference that make wireless transmission a challenging task. In the secondpart of this chapter, we survey the cooperative diversity and its derived singlerelay selection techniques that effectively mitigates the channel impairments. Themajority of cooperative techniques are physical layer oriented and few of theminteract with the MAC layer.

Therefore, we think that a layered approach in wireless sensor networks, whereeach layer stack is unaware of the operation of other layers, eliminates the benefitsof joint optimization across protocol layers. Hence, a joint cross-layer design


between the network and MAC layers on the one hand, and a node cooperationmechanism on the other, is necessary to improve the overall network performance.With this crosslayer approach, we exploit the synergies at the different layers whilesatisfying the network resource constraints. As stated at the introduction of thepresent work, our goal is to integrate the network, MAC and physical layers suchthat the network layer will take advantage of the broadcast nature of wirelesstransmissions to send the packet, the MAC layer will provide us the forwardingnode with respect to a predefined metric and the physical layer will propose thereliability in transmission offered by the cooperative communications.

Chapter 3Models and tools

3.1 Introduction

In the precedent chapter, we have presented the state of art related to thegeographic routing and cooperative communications. These works are based onsome models and assumptions that will be described in this section. You can skipthis part if you are familiarized with the theory, otherwise, we advise you to readit carefully to have a complete overview of how the protocols work.

We organize this chapter in two parts. We start describing the network modelstogether with some graph concepts, then, we describe some common propagationmodels, finally, in the second part we describe the tools used to simulate andcompare the contributed protocols and algorithms.

Before describing in details the models, we present a general overview of asensor network. Let consider a wireless sensor network with a set V of n nodesdistributed in an area, generally a two dimension plane. The sensor nodes arecomposed of one or several sensor units and one omnidirectional antenna. Thenode transmission range is considered as a disk centered at the sensor node witha radius equivalent to the transmission power. A lot of works, consider in generalthe transmission range as a normalized value to the unity. In addition, in somescenarios, the sensor nodes have a Global Positioning System (GPS) receiver,which provides the location information to the node. If GPS devices are notpresent, some alternative methods are used as the relative and virtual coordinatesto embed the node in a graph. The routing protocols working with locationinformation apply some techniques and concepts based on the computationalgeometry to solve some issues in the route construction, thus, the geometricmodels described in the next sections belong to the computational geometry field.

51

52 Chapter 3. Models and tools

3.2 Network models

The previous description of a wireless sensor network corresponds to the UnitDisk Graph (UDG) model, where an edge between two nodes u and v exists ifthe Euclidean distance between the nodes is at most 1. The Euclidean distancecorrespond to the radius of the disk in the plane (observe in Fig. 3.1 the UDGmodel and in Fig. 3.2 a WSN modeled as UDG graph). The network graphat a given time may be static or dynamic, depending on the nodes position,transmission power, mobility patterns and node failures.

Figure 3.1: Unit Disk Graph

Figure 3.2: Unit Disk Graph network

The UDG graph is the most common model used to study the wireless net-works due to its simplicity. However, it is known that this model derived fromwired networks might represent the wireless network inaccurately as it ignores

3.2. Network models 53

several wireless properties. Compared to wired networks, the WSNs need a spe-cial treatment derived from their properties and limitations. These networks arebattery powered with a limited memory. Therefore, the challenge is bigger whentrying to design an efficient routing protocol that saves the resources consump-tion.

(a) Gabriel Graph

(b) Gabriel Graph network

Figure 3.3: Gabriel Graph model and network

In fact, many routing protocols are based on some geometric properties ofthe graph to operate effectively and efficiently. This is the case of the geographicrouting protocols (also known as position-based routing) that are based on theright hand routing such as Greedy Perimeter Stateless Routing (GPSR), GreedyFace Greedy (GFG), Adaptive Face Routing (AFR), Greedy Other Adaptive FaceRouting (GOAFR+), etc. These protocols, ask the network topology to be planarto guarantee the message delivery, and specially avoid the path loops. A planar


graph is one that can be drawn on a plane in such a way that there are no ”edgecrossings,” i.e. edges intersect only at their common vertices. This is difficultto get in real networks, and obviously, the edges may cross when the networkis modeled by the unit disk graph (UDG does not belong to the class of planargraph).

(a) Relative Neighbor-hood Graph

(b) Relative Neighborhoodnetwork

Figure 3.4: RNG Graph model and network

The alternative to this issue is to design a network topology that builds asubgraph of the UDG so that it can be constructed and updated efficiently, andwith some special properties such as planarity, bounded node degree, low-stretchfactor that provides the basis platform to build a localized routing scheme withguaranteed performances. We briefly describe some planar and connected sub-graph derived from the UDG model.

The Minimum Spanning Tree of G denoted MST (G), is the tree belonging toE that connects all nodes and whose total edge length is minimized. MST (G) is

3.2. Network models 55

obviously one of the sparsest connected subgraphs, but its stretch factor can beas large as n− 1 and its construction is centralized.

A Gabriel Graph GG(G), is a graph that connects a set of points in theEuclidean plane. Two points u and v are connected by an edge in the GabrielGraph whenever the circle having line segment uv as its diameter contains noother points from the given point set. More generally, in any dimension, theGabriel graph connects any two points forming the endpoints of the diameter ofan empty sphere. See in Fig. 3.3(a) the GG(G) model and in Fig. 3.3(b) a WSNmodeled as GG(G) graph.

(a) Delaunay trian-gulation Graph

(b) Delaunay Graph

Figure 3.5: Delaunay triangulation model and Delaunay network

The Relative Neighborhood Graph, denoted by RNG(G), is a geometric con-cept proposed by Toussaint [Toussaint 80]. It connects all edges uv ∈ E such thatthere is no point w ∈ V with edges uw and wv in E satisfying ‖ uw ‖<‖ uv ‖and ‖ wv ‖<‖ uv ‖. See Fig. 3.4(a) for an illustration of the RNG(G) and Fig.3.4(b) to see a WSN modeled as RNG(G) graph.


The Gabriel graph was used as a planar subgraph in the GFG and the GPSRprotocols when the protocols work at the recovery mode so as to guarantee thepacket delivery. The Relative neighborhood graph was used for efficient broad-casting in [Seddigh 01] in order to minimize the number of retransmissions in aone-to-one broadcasting model.

The Delaunay Triangulation and its dual, the Voronoi Diagram [Aurenhammer 91],are also two geometric structures used in the conception of some ad hoc routingprotocols. A triangulation of V is a Delaunay triangulation, denoted by Del(V ),if the circumcircle of each of its triangles does not contain any other nodes ofV in its interior. A triangle is called the Delaunay triangle if its circumcircle isempty of nodes of V . The Voronoi region, denoted by V or(v), of a node v ∈ V isthe collection of two dimensional points such that every point is closer to v thanto any other node of V . The Voronoi diagram for V , denoted by V or(V ), is theunion of all Voronoi regions V or(v), where v ∈ V .

Other geometric structures have been proposed that may be applied to geo-graphic routing protocols such as the Yao Graph [Yao 82] where a parameter k(for some integer k ≥ 6), denoted Y G(G), is defined as follows: at each nodeu, divide the plane into k equally separated cones centered at u; then, connectu to its closest neighbor within each cone. In [Li 02], the authors proposed theUnit Delaunay Triangulation UDel(V ) where given a set of points V , UDel(V )is the resulting graph of removing all edges of Del(V ) that are longer than oneunit (UDel(V ) = Del(V ) ∩ UDG(V )). Other interesting alternatives are theRestricted Delaunay Graph (RDG) [Gao 05] and the Local Minimum SpanningTree (LMST) [Li 03].

3.2.1 Discussion

To create valid routing schemes in wireless networks as ad-hoc and sensornetworks, it is necessary to describe the physical relationship between nodes.Here, geometry seems to be the natural tool, as the devices exist, more or less, onthe plane. Furthermore, graphs seem to be a natural representation for networks,and geometric graphs are even more suited for the task. Researchers first made useof the Unit Distance Graph, but eventually that model proved to be insufficient.So other geometric structures were developed to capture the relationships betweennodes more accurately. From these structures, we must attempt to find subgraphsof the graphs that will serve in a routing scheme.

3.3 Power-attenuation model

A critical issue in WSN, is the energy consumption of the nodes that impactthe network lifetime. Each node is composed by transmission, reception and

3.4. Radio propagation models 57

processing units. When a node transmit some data, the involved nodes consumesome energy resources that can be estimated by the sum of three parts. First,the source node consume some resources to prepare the signal. Second, in themost common power-attenuation model, the power needed to establish a linkbetween u and v is ‖uv‖β, where ‖uv‖ represents the Euclidean distance betweenthe nodes, and β is a constant between 2 and 5 dependent on the transmissionenvironment, typically called path loss constant. Finally, when a node receivesthe signal, it needs to consume some power to receive, store, and precess thatsignal. For simplicity, a lot of research works consider it as a constant for all thenodes in the network. Thus we can say that, the power cost p(e) of a link e = uvis the power consumed to transmit a signal from u to v.

3.4 Radio propagation models

The transmitted signal over wireless channels are subject to the physical envi-ronment that modifies the originally transmitted signals at the receiver, introduc-ing some uncertainty about the original encoded and modulated data, resulting inerrors during the transmission. Thus, the wireless signal impacts significativelythe protocols performance so, the networks simulator should take it into con-sideration and use some radio propagation model to obtain information like thesignal strength, attenuation, interference that affect the simulation development.Several models have been proposed which differ basically in their complexity andaccuracy to model the wireless signals.

The free space model is one of the most widely used. It is simple as it consideran isotropic transmission range that allows to behave like a unit disk graph. Thefree-space model consider a propagation through a clear line of sight between thetransmitter and the receiver, and, where the signals are subject to a distancedependent loss of power (path loss). Friis derived an equation to compute thereceived signal power Pr at a distance d from the transmitter.

Pr(d) =PtGtGrλ2

(4π)2d2L

=PtGtGrλ2

(4π)2d20L

.

(d0

d

)2

= Pr(d0).

(d0

d

)2

(3.1)

Where Pt is the power of the transmitted signal, λ correspond to the wave-length, Gt and Gr are the antenna gains of the transmitter and receiver respec-


tively. L ≥ 1 represents the system loss induced by the transmit/receive circuityand generally is set to 1. The transmission radius r is the distance d at which thepower of the received signal Pr(d) equals the receiver sensitivity (minimum re-quired power to receive and decode successfully a packet. Thus, the nodes withinthe transmission range r receive all the packets, and if the distance d > r, theyare not able to receive the packets at all. d0 is the far-field distance, which is areference distance, generally 1 meter for short range systems like wireless datanetworks.

Since the free-space model is rarely found in practical cases, because the mul-tipath propagation due to reflection from the ground was omitted. With thisphenomena present in wireless transmissions, it is likely that not only a singlebut multiple copies for the same signal reach the receiver over multiple paths withdifferent characteristics. Thus, the model was slightly generalized to:

Pr(d) = Pr(d0).

(d0

d

)γ(3.2)

where γ represents the path-loss exponent, which varies between 2 and 5 ac-cording to the environment (free-space and shadowed and obstructed in-buildingsscenarios). The path loss is defined as the ratio of the radiated power to thereceived power, using Eq.3.2, it can be expressed in decibel in the following equa-tion.

PL(d)[dB] = PL(d0)[dB] + 10γlog10

(d

d0

), (3.3)

which is commonly known as the log-distance path loss model. PL(d)[dB] isthe path loss at the reference distance.

Another model, considered as an extension to the log-distance model was pro-posed in order to improve the accuracy of the model by accounting the presenceof obstacles into account as a shadow effect: the log-normal fading model. Here,the shadow effect is considered as a zero-mean Gaussian random variable Xσ,with a standard deviation σ. Thus, the received power in dB can be expressed asfollows:

Allowable path loss PL(d) = Pathloss+ shadow effect

PL(d)[dB] = PL(d0)[dB] + 10γlog10

(d

d0

)+Xσ (3.4)

3.5. Mobility models 59

3.4.1 Discussion

From the propagation models, we can note that the received power at thenodes depends on the frequency, and on the distance between transmitter andreceiver, as well as the reflection, diffraction, scattering phenomena. Thus, awireless routing protocol design should pay a lot of attention on the influencesgenerated by physical layer during a communication process.

3.5 Mobility models

In some sensor networks, nodes can move (e.g., assisted living through wear-able sensors, tracking, etc.) as well as the data collectors. In this chapter weillustrate the most important mobility models that have a great impact on theperformance of routing protocols in wireless sensor networks such as the RandomWaypoint (RWP) and the Brownian model.

Random waypoint model, is the most commonly used mobility model for adhoc networks and sensor networks. The model has been introduced in [Johnson 96],where, each node chooses uniformly at random a destination point (the waypoint)within the deployment region R, and moves toward it along a straight line. Nodevelocity is chosen uniformly at random in the interval [vmin, vmax], where vminand vmax are the minimum and maximum node velocities. When the node ar-rives at destination, it remains stationary for a predefined pause time, and thenstarts moving again according to the same pattern.

RWP represents an individual movement of the nodes with a obstacle-freescenario, meaning that, each node moves independent of each other, and it canmay move in any subregion of R (obstacle-free). The RWP model has also beengeneralized to slightly more realistic model by [Bettstetter 03]. The authors,is extended the RWP model by allowing nodes to choose pause times from anarbitrary probability distribution. Furthermore, a random fraction of the networknodes remains stationary for the entire simulation time. In Fig. 3.6, we plottedmultiples nodes with a RWP behavior.

Brownian motion model has a different vision of RWP where the mobility lookslike intentional movement. This model resembles to a non intentional motion.For this reason, sometimes the model is called drunkard-like model. In Brownianmotion, the position of a node at a given time step depends (probabilistically)on the node position at the previous step. In particular, no explicit modeling ofmovement direction and velocity is used in this model. In [Santi 03], the Brownianmobility is modeled with pstat, pmove and m parameters. The pstat representsthe probability that a node remains stationary for the entire simulation time. Thepmove parameter, is the probability that a node moves at a given time step. Andthe m parameter models the velocity: if a node is moving at step i, its position


Figure 3.6: The Random Waypoint Model behavior, where each color representsthe movement of a node

at step i + 1 is chosen uniformly at random in the square or side 2m centeredat the current node position. We plotted in Fig. 3.7 a simulation a single nodebehaving the Brownian motion model

Figure 3.7: Brownian motion model in a single node

3.6 Simulation tools

Throughout the thesis, various tools were used to simulate and analyze theperformance of the contributed protocols and algorithms. In this section, webriefly describe those tools and indicate why we elected them.

3.6. Simulation tools 61

OMNeT++ [Omnetpp ] simulator along with the Mixim [MiXiM ]frameworkwere used to simulate and analyze the geographic routing protocols at the packet-level. The simulator was developped by Andras Varga from the Department ofTelecommunications (BME-HIT) of the Technical University of Budapest, Hun-gary. OMNeT++ is a discrete event simulation environment with an extensi-ble, modular, component-based C++ simulation library and framework, with anEclipse-based IDE and a graphical runtime environment. Domain-specific func-tionality (support for simulation of communication networks, queuing networks,performance evaluation, etc.) is provided by model frameworks, developed asindependent projects. Its modular structure, its scalability with respect to NS-2,its possibility to run simulations under various user interfaces such as graphi-cal, animating user interfaces are highly useful for demonstration and debuggingpurposes, and command-line user interfaces for batch execution, motivated us totake it as our simulation environment.

Although OMNeT++ provides a powerful and clear simulation framework, itlacks of direct support and a concise modeling chain for wireless communication.Both is provided by MiXiM. MiXiM joins and extends several existing simulationframeworks developed for wireless and mobile simulations in OMNeT++.

Matlab [Matlab ], as we explained above, Omnet++ is a network simulatorat the packet-level. To study the behavior of our Geographic Cooperative protocol(CoopGeo) contribution, we needed to simulate it at bit streams level. In principleits implementation in Omnet and Mixim is possible, but it would carry a lot ofcomplexity, work, and, it would still ask for making some abstraction at thephysical layer. Therefore, we decided to use a different tool, for instance Matlab.With Matlab, we successfully simulated different modulation types as requiredby our contribution.

Chapter 4CoopGeo: A CooperativeGeographic Routing Protocol

In chapters 1 and 2, we presented the Beaconless Geographic Routing pro-tocols as an enhancement to the traditional geographic routing, their vision isbased on a cross-layer design which mixes the network and MAC layer. In fact,this beaconless protocols can not profit from all the bandwidth available to trans-late it to a good achievable throughput as it lacks of a broad wireless vision: itdoes not exploit the broadcast nature of the wireless transmissions. Thus, thecentral object of this chapter is how adapt this beaconless vision to be sensible tothe wireless channel, thus, we propose a system that makes the sensor networkcommunications reliable in terms of end to end delivery, and thus, to cope theinfluence of attenuation due to fading and shadowing during the communicationprocess.

We have focused our attention at the cooperative relaying, which has beenproposed as a promising transmission technique that effectively creates spatial di-versity through the cooperation among distributed nodes. However, we remarkedthat to achieve efficient communications while gaining full benefits from nodescooperation, more interactions at higher layers of the protocol stack, in partic-ular the MAC (Medium Access Control) and network layers, are indispensablyrequired. This is ignored in most existing cooperative relaying works that mainlyfocus on physical-layer cooperative techniques. In this chapter, we describe ourfirst contribution that proposes a cross-layer framework involving two levels ofjoint design—one is a joint MAC-Network cross-layer design for forwarder selec-tion (or termed routing), and the other MAC-Physical for relay selection. Basedon geographical information of nodes and contention processes, the proposedcross-layer protocol, CoopGeo, aims at providing an efficient approach to select

63

64 Chapter 4. CoopGeo: A Cooperative Geographic Routing Protocol

the next hops along the communication path as well as the optimal relay for eachcooperative hop.

The chapter is organized as follows. We start by presenting the context ofcooperative communications and geographic routing. The network model of co-operative multi-hop networks and the problem statement are described in section4.2. Section C.1.2 details CoopGeo, i.e., the proposed cross-layer design for co-operative wireless sensor networks, in which beaconless geographic routing andrelay selection, along with a feasible protocol, are included. In section 4.4, wegive the numerical simulation results for CoopGeo and evaluate its performanceby comparing with an existing protocol.

4.1 Introduction

We stated in chapter 2 that by taking advantage of the broadcast nature ofthe wireless medium, neighbors overhearing data packets can assist the ongo-ing transmission. Such resource sharing (e.g. power, antennas, etc.) achieved bynodes cooperation is a fundamental idea of cooperative communications. In otherwords, the number of degrees of freedom in wireless systems can be effectivelyincreased by enabling collaboration between network nodes. Most attractively,without the requirement of equipping wireless terminals with multiple antennasto construct a multiple-input multiple-output (MIMO) system, cooperative tech-niques break the limitations of the physical size and hardware complexity presentin wireless sensor networks.

Most existing work on cooperative techniques focuses on physical-layer co-operation, where various diversity-oriented signaling strategies have been pro-posed and further demonstrated on the basis of information theory [Laneman 04,Sadek 07, Herhold 04]. However, to achieve efficient communications while gain-ing full benefits from nodes cooperation, more interactions at higher layers ofthe protocol stack, in particular the MAC (Medium Access Control) and networklayers, are indispensably required. Furthermore, an efficient cooperation-basedMAC (or cooperative MAC) scheme should be not only payload-oriented but alsochannel-adaptive to improve the network throughput and diversity gain simulta-neously; otherwise, an inefficient MAC scheme may even make cooperation gaindisappeared [Sanchez 09].

Two major questions related to cooperative MAC design need to be answered:1) when to cooperate? 2) whom to cooperate with and how to do selection? Forthe first question, intuitively there is no need to do cooperation if the direct link isof high quality. In addition, cooperation inevitably introduces inefficiency in somedegree due to extra protocol overhead and limited payload length. Therefore acooperative MAC protocol should be carefully designed to prevent unnecessary

4.1. Introduction 65

cooperation. The second question addresses the typical relay selection problem.There may exist a group of available relays around the source; however, some arebeneficial and some not. How to find the optimal one(s) efficiently and effectivelyis of vital importance to a practical MAC protocol.

Most relay selection schemes focus on the design of enhancing system reliabil-ity in a centralized manner [Yi 08], neglectful of the needs of overhead producedby nodes coordination, unmindful of the feasibility of capturing a lot of chan-nel state information (CSI) among nodes, unsuitable for being used in resource-constrained networks. On the contrary, to the best of our knowledge, there isvery limited study in the literature on joint MAC-physical layer design for relayselection [Liu 05, Shan 08, Wang 09c]. In [Liu 05], a cooperative MAC proto-col called CoopMAC is proposed to alleviate the throughput hindrance causedby low-date-rate nodes. The main idea is to allow a high-data-rate node help-ing the data delivery through two-hop transmission. In [Shan 08], the authorsproposed a busy-tone-based cross-layer cooperative MAC (CTBTMA) protocol,where busy tones are utilized to solve collisions in a cooperation scenario andaddress the optimal relay selection problem. In [Wang 09c], an efficient relayselection scheme based on geographical information is proposed. By jointly com-bining the source-relay and relay-destination distances, the optimal relay offeringthe best cooperative link could be efficiently determined. However, the selectionprocess proposed by [Wang 09c] requires a central controller to decide which re-lay is most helpful, leading to more overhead and power consumption. One goalof the contribution, is to present a distributed relay selection protocol based on[Wang 09c], with MAC-physical cross-layer design.

Likewise, in view of the interaction between the MAC and network layers,we also consider routing issues in this contribution since a properly designedMAC protocol can facilitate routing process at the network layer. In particular,we incorporate the BeaconLess Geographic Routing (BLGR) [Heissenbuttel 03,Fubler 03, Zorzi 03, Blum 03, Sanchez 07]. BLGR is one of the most efficientand scalable routing solutions for wireless ad hoc and sensor networks. The keyadvantage of BLGR is that it needs neither prior knowledge of network topologyfor making a route decision nor the periodic exchange of control messages (i.e.,beacons) for acquiring neighbors’ geographic locations. A current node can makeits own routing decisions by using local information. In general, a BLGR protocolcomprises two operating phases: forwarding phase and recovery phase. In theforwarding phase, routing decisions are made according to the greedy mechanism,a neighbor closet to the destination is chosen as the next hop of a current node.Greedy forwarding, however, fails when reaching a local minimum, i.e., a currentnode that has no neighbor closer to the destination. In this case, the recoverymode based on the well-known face routing algorithm is triggered to find anotherpath to the destination.


It is worth noting that BLGR at the network layer is usually coupled withMAC protocols to offer better network throughput and preserve advantageousproperties such as localized operation and high scalability. A paradigm of network-MAC cross-layered BLGR protocol is as follows: through a contention processat the MAC layer, each candidate forwarder sets a contention timer dependingon the progress to the destination such that the optimal candidate (closest tothe destination in greedy sense) responds first, as a result of time-out. Hence,cross-layer design between the network and MAC layers is quite significant. In[Sanchez 07], Sanchez et al proposed a cross-layered BLGR protocol called BOSS,which uses a three-way (DATA/RESPONSE/SELECTION) handshake and anarea-based timer-assignment function to reduce collisions among responses dur-ing the forwarder selection phase. However, when operating in recovery mode,BOSS performs face routing by requiring the exchange of complete neighborhoodinformation. To avoid this drawback, we present a fully beaconless protocol thatdoes not require beacons in both the greedy forwarding and recovery modes.

Above, we have introduced the roles of interactions between the MAC andphysical layers and between the network and MAC layers, in a cooperative sce-nario. In this contribution, we aim at the integration of the network (NWK),MAC, and physical (PHY) layers as cross-layer design so as to enhance overallsystem performance. Two issues, routing and relay selection, are the two chiefconsiderations. The proposed novel cross-layer framework, called CoopGeo, con-sists of two joint cross-layer designs, a joint network-MAC design for next hopselection and a joint MAC-physical design for relay selection. In particular, boththe routing and relay selection solutions in CoopGeo are geographic protocolsusing contention-based selection processes, providing a strongly practical multi-layer integration for cooperative networks.

The contributions of CoopGeo are:• It proposes a distributed MAC-PHY cross-layer design for relay selection

based on the geographic approach [Wang 09c], where the best relay is chosenas the one that provides the minimum average point-to-point SER.• It presents a fully beaconless approach to geographic routing with a MAC-

NWK cross-layer design, where both the greedy forwarding and recoverymodes are executed without periodic exchange of beacons and completeneighborhood information.• From networking to communications, the proposed protocol CoopGeo pro-

vides a comprehensive integration across the most critical protocol layers,including NWK, MAC, and PHY, to achieve a highly-efficient communica-tion.• Based on the use of geographic information and contention processes, the

framework of CoopGeo that supports localized operation as well as high

4.2. Network Model and Problem Statement 67

(a)

(b)

Figure 4.1: (a) Cooperative multihop sensor network model (b) Direct and coop-erative modes for each hop

scalability is considerably practical for cooperative wireless ad hoc and sen-sor networks.

4.2 Network Model and Problem Statement

4.2.1 Network Model

To design the cross-layer framework, we consider a wireless sensor network ofk nodes randomly deployed in an area, represented as a dynamic graph G(V,E),where V = {v1, v2, . . . , vk} is a finite set of nodes and E = {e1, e2, . . . , el} a finiteset of links between the nodes. We denote a subset N(vi) ⊂ V , i = 1, . . . , k, asthe neighborhood of the node vi, defined as those nodes within the radio rangeof vi.

Fig. C.1(a) depicts the wireless sensor network model, in which the source


S sends its data to the destination D in a multihop manner. In this figure,the dashed circle centered at S illustrates the radio range of S, and so on. Atthe beginning of every data transmission, S broadcasts the data to its neighborsN(S). One of these neighbors N(S) is chosen as the next hop through a for-warder selection process, denoted as F1. Two transmission modes, namely directand cooperative modes, are considered to operate in each hop. In the directmode, an point-to-point communication is performed by direct transmission; inthe cooperative mode, it is done by cooperative relaying. The cooperative modeoperates only when F1 cannot correctly decode the data from S. After havinga correct version of the data packet, F1 acts as the source node and repeats thesame procedure, and so on until the data packet reaches the destination D.

Since the multihop transmission is realized by concatenating multiple single-hop schemes, as shown in Fig. C.1(a), for convenience of notations we denoteS as the current source in a current hop and F the forwarder, also called thenext hop or the intermediate node in this chapter. In addition, we represent Ri,i = 1, . . . , |N(S)|, as the candidate relays of S, one of which is going to cooperatewith S whenever needed. Fig. C.1(b) illustrates the transmission schemes fordirect and cooperative modes respectively, where the only difference betweenthem is that F also receives the data from R in the cooperative mode, but not inthe direct mode. In the following we introduce the signal models for the directand cooperative transmission modes, respectively.

In the direct mode, S broadcasts its symbol x with transmission power P ,where the average power of x is normalized to unity. The received signals at Fcan be expressed as

yS,F =√PhS,Fx+ nS,F , (4.1)

where hS,F is the channel fading coefficient from S to F , modeled as hS,F ∼CN(0, σ2

S,F ); the additive noise term nS,F is a circularly symmetric complex Gaus-sian random variable, assumed as nS,F ∼ CN(0, N0). Without loss of generality,we assume the noise terms have equal variance with N0 = 1.

For the cooperative mode, it applies a two-phase decode-and-forward (DF)strategy with single-relay selection, described as follows. In the first phase, Sbroadcasts its symbol x with transmission power Px while the next hop F and aselected relay R (through a relay selection process) listen. The received signalsat F and R can be respectively expressed as

yS,F =√PxhS,Fx+ nS,F , (4.2)

yS,R =√PxhS,Rx+ nS,R, (4.3)

where hS,R is the channel fading coefficient from S to R, modeled as hS,R ∼CN(0, σ2

S,R); the additive noise terms nS,R is a circularly symmetric complex

4.2. Network Model and Problem Statement 69

Gaussian random variable, assumed as nS,R ∼ CN(0, N0). In the second phase,with the simple adaptive DF strategy [Herhold 04], the selected relay decideswhether to forward the decoded symbol to the next hop. If the relay is able todecode the transmitted symbol correctly, it forwards the decoded symbol withidentical transmission power Px to the next hop, and if not, it remains idle. Inpractical scenarios, this ’adaptive’ mechanism can be achieved based on an SNRthreshold. If the SNR at the relay is greater than a certain threshold, the relayforwards; otherwise, it remains idle. Define the indicator function IR as follows:

IR =

{1, if R decodes the transmitted symbol correctly,0, otherwise.

(4.4)

Then, the received signals at the the next hop in the second phase can be writtenas

yR,F =√PxIRhR,Fx+ nR,F , (4.5)

where hR,F denotes the channel fading coefficient from R to F , modeled as hR,F ∼CN(0, σ2

R,F ), and nR,F denotes AWGN, nR,F ∼ CN(0, N0). Finally, the next hopcoherently combines the received signals from the current source and the selectedrelay, i.e., yS,F and yR,F , by using a maximum ratio combining (MRC)

yF =√Pxh

∗S,FyS,F +

√PxIRh

∗R,FyR,F . (4.6)

Consequently, the decoded symbol x at the next hop is given by

x = arg minx∈A

|yF − Px(|hS,F |2 + IR|hR,F |2)x|2, (4.7)

where |A| = Θ denotes the cardinality of Θ-ary constellation.By invoking the performance analysis in [Su 08], the resulting symbol error

rate (SER) at the next hop can be expressed as

Ps ≈4N2

0

b2P 2xσ

2S,F

(A2

σ2S,R

+B

σ2R,F

), (4.8)

which is a tight approximation in a high SNR regime, where ifM -PSK modulationis used, b = sin2(π/M), and

A =M − 1

2M+sin(2π

M)

4π,B =

3(M − 1)

8M+sin(2π

M)

4π−sin(4π

M)

32π; (4.9)


Figure 4.2: Area division for CoopGeo routing. F1 and F2 are sub-area 0 and 1of PPA respectively, whereas F3 and F4 are sub-area 4 and 5 of NPA respectively.

if M -QAM, b = 3/2(M − 1), and

A =M − 1

2M+

(1− 1/√M)2

π,B =

3(M − 1)

8M+

(1− 1/√M)

2

π. (4.10)

Moreover, we make the following assumptions in the network model: 1) eachnode has a single antenna operating over frequency-flat fading channels and canonly either transmit or receive data at any time slot; 2) The network is dynamicand the network topology, including the cardinality of a node’s neighborhood,the location of nodes, and the linkage between nodes, changes over time dueto wireless environments, duty circles, and node failures, etc.; 3) each node isaware of its own location; 4) In addition to itself’s location, the source knowsthe location of the destination, and so does any intermediate node; 5) All thenetwork nodes are homogeneous, and each could become a source, a relay, or aforwarder.

4.2.2 Problem Statement

In considering how cross-layer design improves network throughput and reli-ability for wireless cooperative multihop sensor networks, the first question thatarises concerns the joint MAC-Network cross-layer routing design. For a networkG(V,E), given a source-destination pair vS, vD ∈ V , the objective of a routingtask is to find a subset of forwarders PF = {vF1 , vF2 , . . . , vFn} ⊂ V that buildsa routing path from vS to vD with successful packet delivery guaranteed. Inparticular, each forwarder in PF is determined locally, within a forwarding areadefined as the radio coverage of the current source, which is divided into a posi-tive progress area (PPA) and a negative progress area (NPA). In both, the PPA

4.3. CoopGeo: A geographic cross-layer protocol for cooperative wireless networks71

Figure 4.3: CoopGeo architecture

and PPA areas, the beaconless greedy forwarding BLGF) and beaconless recoveryforwarding (BLRF) phases are applied, respectively (as shown in Fig. C.2).

The second question that this study addresses concerns the joint MAC-PHYcross-layer relay selection design. The aim of CoopGeo relay selection is to finda subset of optimal relay nodes PR = {vR1 , vR2 , . . . , vRm} ⊂ V \ PF to enhancethe network reliability, where each optimal relay vRi that minimizes the averagepoint-to-point SER for each cooperative hop is locally selected within a relay areadefined as a reuleaux triangle.

One design goal of CoopGeo is to develop a fully beaconless approach togeographic routing that does not rely on periodic exchange of beacons as wellas complete neighborhood information. Therefore, forwarder and relay selectionuse a local contention process based on geographical information and area-basedtimers. A specified interval of time Tmax is assigned to each selection process.

By tackling the above issues, we contemplate a feasible cross-layer protocolthat comprehensively integrates the NWK, MAC, and PHY layers to achieve ahighly-efficient communication. In the following section we detail the frameworkof the proposed cross-layer design.


4.3 CoopGeo: A geographic cross-layer proto-

col for cooperative wireless networks

CoopGeo, in general, performs two tasks in wireless cooperative multihopsensor networks: routing and relay selection (see CoopGeo architecture in Fig.4.3). As described above, the routing process works in two phases, i.e. BLGFand BLRF. Both phases share equally a Tmax interval of time where the forwarderselection is executed. The first half of the Tmax period is allocated to the BLGFphase and the second half to the BLRF phase.

In the BLGF phase, a next hop that provides maximal progress toward thedestination is selected through a timer-based contention process. When failing tofind a next hop in the BLGF phase, the routing process enters transparently tothe BLRF phase and applies face routing by using graph planarization along witha select-and-protest principle. Cooperative relaying is required after the routingtask, whenever the selected next hop decodes the data packet erroneously. Inthis case, CoopGeo starts out to execute the relay selection task within anotherinterval of time Tmax, selecting an optimal relay that offers the best cooperativelink between the current source and the next hop.

Fig. C.1(a) gives an example for both the routing and relay selection in Coop-Geo. The nodes competing in the BLGF phase are those located in PPA, i.e.,X1, X2, R1, and F1. Those located in NPA, i.e., W1, . . . ,W4, are considered tocompete in the BLRF phase. The node F1 is selected as the forwarder node,where the data transmission between the source S and the forwarder F1 is car-ried out through a direct or cooperative transmission. In the case a cooperativetransmission is used, the candidate relays with respect to the transmitter-receiverpair (S, F1) that participate during the relay selection process are those withinthe relaying area (that will be defined later), including R1 and X1. In this figureR1 is selected as the optimal relay node of the cooperative transmission.

4.3.1 Beaconless Greedy Forwarding (BLGF)

At the beginning of a data transmission, S triggers the BLGF phase of therouting process by broadcasting its data to the neighborhood and then waits forthe best next hop’s response during the first half of the Tmax time. Duringthis period, the neighborhood compete to forward the message by setting theircontention-based timers (TCBF ), as explained in Section 4.3.1.1. When the bestforwarder node is selected due to its timer expiration, it sends a Clear-To-Forward(CTF) message to S, then, the other candidates overhearing this message sup-press their running timers and delete the data received from S. Since some can-didates situated at the forwarding area may be unable to hear the CTF message,the hidden terminal problem could appear. To prevent it, S broadcast a warn-


ing message (SELECT) to indicate that a forwarder node has been found. Thehidden candidates overhearing it, will suppress their timer and the data packet.Immediately, F send an acknowledgement (ACK) to S and thereafter, it acts asthe source and repeat the process hop-by-hop until the data is delivered to thefinal destination D.

4.3.1.1 Geographic contention-based forwarder selection (TCBF )

To implement the BLGF phase, we base the timers settings on the metricproposed in [Sanchez 07], which applies an area-based assignment function. Fig.C.2 depicts, as mentioned above, two areas, PPA and NPA, that divide the radiocoverage of a current source, both of which are further divided into sub-areascalled Common Sub-Areas (CSAs) in order to avoid collisions during the con-tention period. Moreover, those candidate nodes situated at the same CSA offersimilar progress toward D, and thus, they have similar TCBF values. Note thatunlike [Sanchez 07], we divide the NPA area by using concentric coronas insteadof slides as used at the PPA area. We will discuss the reason at the BLRF section.

The timer setting for each candidate node is given as follows. First, eachcandidate node situated in PPA identifies which CSA group it belongs to byusing the following equation:

CSAPPA =⌊NSA× r − (dS,D − dFi,D)

2r

⌋(4.11)

where NSA is a predefined even number of sub-areas to divide the coveragearea, r is the transmission range which is equal to the largest progress, and(dS,D − dFi,D) represents the candidate progress to the destination (candidateslocated in NPA use Eq. (C.12).

Next, given CSAPPA, hereafter called CSA, each candidate calculates its TCBFtimer according to:

TCBF =(CSA× Tmax

NSA

)+ rand

( TmaxNSA

)(4.12)

where Tmax represents the maximum delay time that the current source S willwait for a next hop’s response, and rand(x) a function obtaining a random valuebetween 0 and x to reduce the collision probability. The TCBF function allocatesthe first half of Tmax to PPA candidates for the BLGF phase and the second halfto the NPA candidates for the BLRF phase.

4.3.2 Beaconless Recovery Forwarding (BLRF)

As introduced before, the BLGF mode may suffer from the local minimumproblem: the packet may be stuck at a node that does not have a neighbor (at


Figure 4.4: Recovery forwarding area is divided in N coronas. Each has a width(√i−√i− 1)r1

PPA) closer to the destination than itself. To solve this problem, the Beacon-less Forwarder Planarization (BFP) algorithm of [Kalosha 08], which guaranteesthe packet delivery is applied at BLRF. BFP reduces the number of messagesexchanged by using the select-and-protest principle. In the select stage, someNPA neighbors are selected to form a planar subgraph according to a contentionfunction, then, in the protest stage, falsely planar edges are removed from thesubgraph. Finally, the traditional face routing algorithm is applied to select theforwarder node.

BFP is implemented at the BLRF phase of CoopGeo as follows, first, thecurrent source detects the local minimum when a Tmax/2 time has passed withoutreceiving any CTF message from any neighbor situated at PPA. Thus, CoopGeoswitches automatically from the BLGF mode to the BLRF mode, so that, BFP isapplied during the second half of Tmax. To accomplish this, the candidate nodessituated at the NPA area determine their CSAs and compute their contentiontimers (TCBF ) that will be used by the BFP algorithm. Once the planar subgraphis build, S send a SELECT message to the node that has been elected forwardernode, which confirms the reception with an acknowledgement.

In [Sanchez 07], the CSAs of NPA are created according to the progress towardthe destination. CoopGeo, by contrast, adopts the distance with respect to thenode that is suffering the local minimum problem, and accordingly, the slides aremodified to concentric coronas.

Thus, The NPA area is divided into n = NSA2

equally sized concentric coronas

(see Fig. 4.4), where the width of the i-th corona is (√i −√i− 1)r1, and r1 is

the radius of the first corona, being calculated with r1 = r/√n. To use the same

terminology as the one used at the BLGF phase, in the following a corona will


Figure 4.5: Nodes vi...vn have the same distance to u. So, each node has to usea random function rand(Tmax

NSA) to decrease the collision probability

be referred as a CSA. To set a contention timer, a candidate F in NPA first findsits CSANPA index by using the following equation:

CSANPA =⌊(√n · dv,u

r

)2⌋+NSA

2(4.13)

With knowledge of their CSANPA index, hereafter called CSA 1, each NPAforwarder candidate determines its contention timer according to (C.11), ratherthan the original equation defined by Kalosha expressed here just as referenceand finally, BFP is applied.

TCBF =d

r· Tmax (4.14)

where d represents the distance from the neighboring node to the node suffer-ing the local minimum problem, r the transmission range and Tmax the maximumdelay affected to a contention timer.

The concentric coronas and the contention delay computed by Eq. (C.11)compared to Kalosha allows that:

1. Nodes located in CSAi broadcast their CTF messages before the nodes inCSAj where i < j.

2. All nodes located in the same CSAi and eventually at the same distance toS (see Fig. 4.5 will reduce their collision probability when sending the CTF

messages because the second term rand(TmaxNSA

)in Eq. (C.11) guarantees

that nodes choose different times to transmit.

1. ”A CSA value at the forwarding selection is a nonnegative integer that falls in [0, NSA−1],where 0 corresponds to the area closest to D and (NSA− 1) to the farthest one”


Figure 4.6: Beaconless recovery messages exchange

It has been proved that face traversal on a planar subgraph for the recoverymode of a geographic routing protocol is loop free and guarantees the packetdelivery [Bose 01]. However, the planar subgraph construction requires a com-plete knowledge of the neighborhood via the beacon exchanges. In contrast, BFPreduces the number of messages exchanged using the selection and the protestphases where the former aims to construct a planar subgraph and the latter gen-erate messages to remove falsely selected neighbors from the planar subgraph.

In this work, we do not explain the BFP algorithm of [Kalosha 08] in detail.Instead, we present an example to illustrate the process in Fig. 4.6 and Fig. 4.7.Let’s consider a scenario where the source S is surrounded by six neighbors whichrespond in the order: F1, F4, and F5 according to their timers defined by (C.11).F2 receives the CTF message from F1 and becomes a hidden node, F3 receivesthe CTF from F4, and F6 receives the CTF from F5. Thus, the hidden nodes areF2, F3 and F6. F2 is located in the proximity region (Gabriel Graph) of F1 andF3 in the proximity region of F4. So, in the protest phase, F2 protests against F1

and F3 protests F4. Thus, S removes the links with violating nodes (node in theproximity region of a node) and obtains a planar subgraph that will be used bythe face routing algorithm to find the next forwarding node.

4.3.3 MAC-PHY Cross-Layered Relay Selection

4.3.3.1 Relay selection criterion based on geographical information

A relay selection criterion based on geographical information, where the bestrelay is determined according to a distance-dependent metric mi that softly com-bines the source-relay and relay-destination distances on each candidate relay isdetermined. The relay selection criterion for each cooperative hop can be ex-pressed as follows,

i∗ = arg mini∈{1,2,...,N}

mi = arg mini∈{1,2,...,N}

A2dpS,Ri +BdpRi,F , (4.15)

where dS,Ri and dRi,F are the distances between the current source and the


Figure 4.7: Beaconless Recovery Forwarding happens at NPA area when theBeaconless Greedy Forwarding fails

i-th relay and between the i-th relay and the next hop, respectively and A and Bare modulation dependent constants that satisfies equations (4.9) and (C.9). Wenote that the best relay selected by the above criterion is the one that providesthe best source-relay-forwarder cooperative link in terms of average SER at thenext hop.

4.3.3.2 Geographic contention-based relay selection

The distributed relay selection is based on nodes geographical information anda contention process. First, each relay acquires two relative distances dS,Ri anddRi,F to calculate its own selection metric according to (C.13). Here the path lossexponent is assumed as a known parameter. For the purpose of decentralization,the relay selection metric mi is encoded in time difference inside a timer-basedelection scheme. The election process starts as soon as each candidate relay over-hears the DATA/CTF packets. Each candidate relay sets its timer proportionalto the selection metric. Once a candidate whose timer expiration happens first, itrelays the data packet to F , and the others candidates cancel their timers after re-ceiving the packet. This contention-based relay selection scheme, which providesa distributed and efficient way to determine the best relay for each cooperativehop, answers a major question about cooperative MAC design, i.e., whom tocooperate with and how to do selection? The metric defined in (C.14) indicatesthe cooperative link quality in terms of average point-to-point SER, dependingon the modulation type and the locations of nodes. We denote xS,xF , and xi asthe locations of the current source, the forwarder, and the i-th candidate relay,respectively.


In addition, we define f as a mapping function that maps a candidate relay’slocation into its relay selection metric (xS and xF are fixed), as in (C.15). Letx∗ be the best placement of a relay, which minimizes the average point-to-pointSER. The optimal point x∗ can be obtained by solving the optimization problem(C.16). Thus, the best relay is the one whose metric is closest to f(x∗).

mi , A2dpS,Ri +BdpRi,F , i = 1, 2, . . . , N, (4.16)

f(xi) = A2 ‖xi − xS‖p +B ‖xi − xF‖p (4.17)

minimize f(xi) = A2 ‖x− xS‖p +B ‖x− xF‖p (4.18)

x∗ =A2xS +BxFA2 +B

(as p = 2) (4.19)

We derive a mapping function M, which scales our metric function f into theinterval [0, 1], where xmax is the point in a set:

M(f(x)) =f(x)− f(x∗)

f(xmax)− f(x∗)(4.20)

Finally, as for the CBF timers, we use the following equation to allocate the timeto each node in the contention-based Relay selection scheme (CBR).

TCBR = Tmax M(f(x)) + rand(2TmaxNSA

)(4.21)

4.3.3.3 Relay selection area

The CoopGeo relay selection process do not use control messages as in theforwarding selection process so as to guarantee that only one node has beenselected as relay, and thus, avoid message duplications or collisions. Besidesoverhearing the relayed message that triggers the contention timers suppressionof the other candidate nodes, we have considered the relay area size as a wayto control these issues. Since the candidates should reside in a defined areawhere relay selection is executed, the relay area is determined by the sourceand forwarder nodes positions. In Fig. 4.8(a) and 4.8(b), two relay areas aredepicted. First, let the set C represent the potential relay nodes situated at therelay area formed by the intersection of the source and forwarder node coverageareas. Second, let the set D represent a relay area shaped by a Reuleaux trianglefrom the source node point of view. In the first case, for any relay candidate xi ∈C, its selection metric is mapped onto this set, where M(f(xi)) ∈ [0, 1]. For theReuleaux triangle, any relay xi and any other possible relay xj have the followingrelationship: ‖xi − xj‖2 ≤ r,∀xi,xj ∈ D, i 6= j where r is the transmission rangeof a node. Hence, from the relay areas depicted in the figure, the Reuleaux


(a) (b)

Figure 4.8: (a) Mapping of the metric on to the set C (b) Mapping of the metricon to the set D for a normalized distance Source(0,0) Destination(1,0)

triangle area is the best suited to be used since all relay candidates can hear toeach other, and accordingly, the hidden relay problem can be effectively avoidedwhich is not the case of the intersection relay area.

4.3.4 CoopGeo in Action

In this section, we depict the behavior of nodes running CoopGeo (cf. Fig.4.9) when a data transmission between a source and a destination is performed.

When the source node S intends to transmits its data to a node D, it checksif the channel is free for a predefined time interval. If this is the case, S broadcastits DATA and starts a TS1 timer. The neighbors of the source node receive thepacket, store it and set up their TCBF timers as defined in Eq. (C.11) so as toparticipate in the forwarding selection process.

The neighbor F ∈ Fi whose timer expires first sends a CTF control message toclaim the forwarding status, then, it initializes a TF1 timer. The other candidateshearing this control message quit the forwarding selection process.

The DATA/CTF handshake carried out by S and F is used to initiate therelay cooperation on demand if it is needed, since within the CTF message, Findicates if relay cooperation is needed in case of error decoding. In this way,the neighbors situated at the relay region formed by S and F that successfullydecoded the DATA previously sent by S start their TCBR timers as defined in Eq.(C.19), so as to participate in the relay selection process.

When S receives the CTF message, it replies with a SELECT message whichconfirms the forwarding status to F , and updates its TS1 timer to the maximumdelay time allowed to receive an ACK from F . Meanwhile, the relay candidatesdecrement their TCTR timers. Thus, when the candidate R ∈ Ri expires itsTCTR timer the first, it becomes the relay node and immediately relay the stored


Figure 4.9: CoopGeo in action

data. The other candidates hearing the transmission notice that another nodehas relayed the data and quit the relay selection.

Consequently, the forwarder node combines the received signals from S andR, decodes the data, and stops its TF1 timer. Then, it sends an acknowledgmentto S and continues the CoopGeo execution toward D.

In addition to TCBF and TCBR timers, two other timers were used: TS1 at thesource and TF1 at forwarder node. At the beginning TS1 represents, the maximumallowed time to find a forwarder node in the direction of D, given by

TS1 = TDATA + TCTF + Tmax (4.22)

Where TDATA and TCTF represent the data and CTF packet transmissiontimes respectively, and, Tmax represents the maximum time interval allowed tothe forwarding selection process. For simplicity, in the equation, we do not expressthe propagation delay.

Since the DATA/CTF handshake represents that a forwarder node F wasselected, TS1 is updated to a value that represents the maximum delay timeallowed to receive an acknowledgement from F and it depends on whether relaycooperation is executed. The updated timer is given by

4.4. Performance Evaluation 81

TS1 =

{TSEL + TACK if no cooperation is needed

TSEL + Tmax + TDATA + TACK otherwise(4.23)

where the first statement includes the transmission time of the SELECT (fromS to F ) and ACK (from F to S) messages; the second statement includes therequired time of the first statement as well as Tmax and TDATA which correspondto the maximum allowed time for the relay selection and the time needed to relaythe packet.

For TF1, the affected value depends on whether the forwarder F correctlydecodes the received data from S, or relay cooperation is executed. For theformer, F listens to the channel and waits for a SELECT message from S, whichcompletes the direct communication mode; for the latter, F waits for the SELECTmessage and DATA relayed, from the source and the relay node, respectively.

TF1 =

{TCTF + TSEL if no cooperation is needed

TCTF + TSEL + Tmax + TDATA otherwise(4.24)

where the first statement allocates to TF1 the time required to transmit theCTF message and the time required to receive a SELECT message from S respec-tively; the second statement adds to the first statement, the maximum allowedtime to select a relay node and the time the relay node needs to send the relayeddata, respectively. Similar to TS1 timer, TF1 does not consider the propagationdelay.

If the timer TS1 of S expires before receiving a CTF or an ACK from F , wehave different possibilities: 1) S could not find a forwarder; 2) F could not receivethe SELECT message from S 3) F could fail; 4) F could not receive the datapacket from R in the cooperative mode. For all these situations, the CoopGeoprotocol is restarted.

Thus, we can see that the two most significant timers are TCBF and TCBR,which are used to select a forwarder F and an optimal relay R in each hop throughcontention mechanisms. The timers TS1 and TF1 just helps to detect a problemduring the CoopGeo execution.

4.4 Performance Evaluation

We first consider a single-hop cooperative relay network with N = 5 availablerelays, deployed in R2. We locate the source and the destination at the coor-dinates of (0, 0) and (1, 0) respectively, and randomly place the relays at thelocation displayed in the column (A) of Table 4.1. We assume that the chan-


Figure 4.10: Comparison of each possible relay selection and random relay selec-tion

nel variances between any two nodes follow σ2i,j ∝ d−pi,j , where p is the path loss

exponent and is taken to be p = 3 in our simulations. The channel variance isnormalized to unity per unit distance. QPSK modulation is used in this simula-tion and the fading channels are assumed sufficiently fast-varying such that thechannel coefficients keep constant only within every symbol interval.

Table 4.1: Relays Locations and The Corresponding Selection Metrics

Relay # (A) Location (B) Selection metric m

R1 (0.32,0.13) 0.1283

R2 (0.08,-0.18) 0.2980

R3 (0.71,0.28) 0.1155

R4 (0.43,0.47) 0.1989

R5 (0.58,-0.07) 0.0691

From the column (A) of Table 4.1, we can determine the distances from eachrelay to the source and the destination, which are known at the source, and thenthe corresponding selection metric for each relay can be determined by using(C.13), given in the column (B) of Table 4.1. According to the selection criterionintroduced in Section III, R5 turns out to be the best relay selection since it hasthe minimum selection metric.

Fig. 4.10 depicts the SER versus SNR performance of the above scenario,


Table 4.2: Simulation Settings

Input Value Input Value

No. of Neighbors 1-20 Tx. Power 25 dBm

Channel Model Rayleigh Average Noise 20 dB

Carrier Frequency 2.412 Ghz Noise Figure 15 dB

Channel Bandwidth 22 Mhz Packet Size 1538 Bytes

Modulation Type QAM No. of Topologies 20000

Constellation Size 4-128 No. of Simu. Trials 2000000

Contention Period 500 µs

where SNR is defined as P/N0, and P is the total transmitted power fixed ineach case. In Fig. 4.10, the performance of direct transmission from the sourceto the destination is provided as a benchmark for a non-cooperation scheme. Fig.4.10 shows that R5 is the best relay since it contributes to the minimum SER atthe destination. Moreover, Fig. 4.10 also reveals that the worse relay (leadingto the worse SER performance) corresponds to the larger selection metric in thecolumn (B) of Table 4.1. In other words, the simulation results are consistentwith the proposed relay selection, that is, the smaller the selection metrics, thebetter the resulting SER performance. Thus, we have demonstrated that by uti-lizing the geographical information, nodes in cooperative networks can efficientlyperform relay selection to improve the SER performance at the destination. Inaddition, we also compare the performance with a possible relay selection ap-proach, named random relay selection, which means that the source randomlyselects a cooperating relay without any information for each transmission. Wesee, in Fig. 4.10, that the performance curve of the random selection scheme liesbetween the best and the worst selection. This is because each relay has the sameopportunities to be selected so that the performance will be averaged over all thedistributed relays.

The next step in our simulation methodology is to evaluate the PHY/MAClayer performance of CoopGeo with Monte-Carlo simulations implemented inMatlab code. We simulated the three lower layer processes, and a summaryof our simulation settings could be found in table C.1. Our results are based on20,000 random generated topologies where all the stations are competing to accessthe channel. We start by solving the two subproblems stated in section 4.2, andonce the forwarder and relay node sets are obtained, we use them to evaluate thepacket error rate, the average transmission probability, the saturated throughput,and some other results varying the input parameters.


4.4.1 Packet Error Rate (PER)

To analyze the PER in both protocols, we start by simulating the protocolswith different Tmax values. Starting from Tmax = 100µs until Tmax = 1000µs, seethe results in Figs. C.3(a) C.3(b).

(a) PER for BOSS (b) PER for CoopGeo

Figure 4.11: PER for BOSS and CoopGeo using Tmax = 100...1000 µs. The curveslocated at the bottom of the figures correspond to minimum value of Tmax = 100and those located in the upper side to the maximum value Tmax = 1000

The two plots allow us to have a global visualization of the protocols behaviorwith respect to the PER. From these results, we have chosen the behavior whenCoopGeo works with Tmax = 500µs as this is the value that presents a goodrelationship between the PER and the delay needed to choose the forwardingand relay nodes. So, in Fig. C.4, we show the average Packet Error Rate of twodifferent protocols, one is for CoopGeo using a cooperative relaying technique andthe other is a BOSS[Sanchez 07] like protocol without cooperative relaying. Thepacket error rate presented in Fig. C.4 includes both the probability of collisioninside different contention periods and the probability of error over the wirelesschannel.

We show that our protocol experienced a lower error rate of 2.5 times lessthan the traditional geographic based routing protocol in the best circumstances.We also notice that the error rate of the two protocols gets closer to each other asa function of the increased number of nodes in the neighborhood. This error rateis a function of the number of nodes and is induced by the collision probabilityinside the different contention periods.


2 4 6 8 10 12 14 16 18 20 220

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Contending nodes

Pac

ket E

rror

Rat

e

BOSSCoopGeo

Figure 4.12: Packet Error Rate for Tmax = 500µs

4.4.2 End to End Transmission Error Probability

To calculate the error rate during a whole transmission, the same way as theprevious analysis, we simulate our contribution with different values for Tmax =100...1000 µs. These simulation can be observed in Figs. 4.13(a) 4.13(b).

Therefore, in Fig. C.5, we show that the average probability of having anerror in the transmission is clearly better in the cooperative case and the rate iseven decreasing as the number of stations present inside the neighborhood grows.This behavior is due to the accurate selection of the relay node when more nodesare present in the neighborhood. We can also notice that CooopGeo experimenta very low transmission error rate that enable us to raise the constellation sizeof the modulation scheme, in order to improve the bandwidth efficiency withoutloosing end to end throughput.

4.4.3 Varying the contention window Tmax

In this test, we investigate the impact of the contention window size Tmax (thatcontrols the delay affected to a contending node when it tries to forward/relaya packet) on the CoopGeo performance. Initially, we simulate our protocol withTmax values from 100µs to 1000µs.

4.4.3.1 CTF-Relayed message Collision Probability

In Fig. C.6(a) we analyse the collision probability suffered by contendingnodes when sending their Contention to Forward and relayed messages according


(a) Error Transmission Rate for BOSS (b) Error Transmission Rate for Coop-Geo

Figure 4.13: Error Transmission Rate (end to end) for BOSS and CoopGeo usingTmax = [100,...,1000] µs

to the Tmax size. The sizes from 500µs to 1000µs are the best suited to be usedby CoopGeo, as the collision probability is similar and stay low in comparisonthe other intervals. We then, analyze the relationship between the normalizedthroughput with cooperative communications vs the CTF-Relayed messages colli-sion probability and observe that we may use a smaller Tmax size without affectingthe performance of the protocol provided that the number of contending nodesbe fewer for the case Tmax = 300µs (cf. Fig. 4.19). If take Tmax = 500µs from theprevious result as reference, we can see that for a smaller saturated throughputrate with respect to Tmax = 300µs we may handle more dense scenarios.

4.4.3.2 Varying the constellation size

Finally, we realized a series of test where we varied the constellation size ofthe protocol (see Figs 4.16 and 4.17). In Fig. C.6(b), we resume the previous re-sults in one figure, showing the saturated throughput of our (NWK/MAC/PHY)CoopGeo and compare it with a traditional geographic NWK/MAC approachsuch as BOSS. We showed that our proposal outperforms the classical schemein terms of saturated throughput, using for this case, different order of the M-Quadrature Amplitude Modulation (MQAM). Due to very low transmission errorrate in the case the cooperative scheme, we are able to raise the size the of con-stellation with respect to the transmission environment without deteriorating theend to throughput.

4.5. Discussion and conclusions 87

2 4 6 8 10 12 14 16 18 20 220

0.05

0.1

0.15

0.2

0.25

Contending nodes

Err

or T

rans

mis

sion

Pro

babi

lity

BOSSCoopGeo

Figure 4.14: End to End Transmission Error Probability for Tmax = 500µs

4.5 Discussion and conclusions

In this chapter, we have presented a cross-layer protocol, CoopGeo, based ongeographic information to effectively integrate the network/MAC/physical layersfor cooperative wireless sensor networks. The proposed CoopGeo provides a jointMAC/routing protocol for forwarder selection as well as a joint MAC/physicalprotocol for relay selection. Both selection methods, are based on geographicalinformation of the nodes, this information is embedded in their metrics whichparticipate in a contention-based mechanism. Simulation results demonstratethat the proposed CoopGeo can work with different densities and achieve bettersystem performances than the existing protocol like BOSS, in terms of packeterror rate, transmission error probability, and saturated throughput. Due to thefact that the forwarder and relay selections take place at the MAC layer, we makeCoopGeo is also reliable and resilient to any change in the network topology andalso very scalable since both decision are local.


2 4 6 8 10 12 14 16 18 20 220

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Contending nodes

Col

lisio

n P

roba

bilit

y

Tmax = 100Tmax = 200Tmax = 300Tmax = 500Tmax = 800Tmax = 1000

Figure 4.15: CTF-Relayed message collision probability when changing Tmax from100µs to 1000µs

0 2 4 6 8 10 12 14 16 18 203

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6x 10

6

Contending nodes in area

Saturation throughput for QAM: 16

DirectCooperative

(a)

0 2 4 6 8 10 12 14 16 18 203

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6x 10

6



DirectCooperative

(b)

Figure 4.16: Saturation throughput for QAM: 16-32

4.5. Discussion and conclusions 89

0 2 4 6 8 10 12 14 16 18 203

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6

4.8x 10

6



DirectCooperative

(a)

0 2 4 6 8 10 12 14 16 18 203

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6

4.8x 10

6



DirectCooperative

(b)

Figure 4.17: Saturation throughput for QAM: 64-128

0 2 4 6 8 10 12 14 16 18 203.2

3.4

3.6

3.8

4

4.2

4.4

4.6

4.8x 10

6

Contending nodes

Sat

urat

ed th

roug

hput

Coop 16QAMCoop 32QAMCoop 64QAMCoop 128QAM

Figure 4.18: CoopGeo Saturated throughput for QAM fom 16-128


2 4 6 8 10 12 14 16 18 20 220.05

0.1

0.15

0.2

0.25

0.3

0.35

0.4

0.45

0.5

Contending nodes

Col

lisio

n pr

obab

ility

and

nor

mal

ized

thro

ughp

ut

Collision prob. for Tmax = 300Sat throughput for Tmax = 300Collision prob. for Tmax = 500Sat throughput for Tmax = 500

Figure 4.19: Normalized saturated throughput and collision probability for Tmax= 300µs and Tmax = 500µs

Chapter 5RACR: Relay-AwareCooperative Routing

In chapter 2 we stated that cooperative communications for wireless networkshas been extensively investigated from the physical layer, then, in chapter 4,we presented a crosslayer framework that provides reliability in terms of packetdelivery guaranteed and a better bandwidth utilization in terms of throughput.Now, in this chapter, our goal is to design an efficient wireless system able tocontrol the resource consumption. To tackle this goal, we present our secondcontribution, a cooperative geographic routing protocol with cross-layer design,named, the Relay-Aware Cooperative Routing (RACR) protocol, which exploitsthe coverage extension property as a result from node cooperation. The RACRprotocol enables a forwarding node offering a certain symbol error rate (SER) toparticipate in the routing process by means of a local route decision that is basedon a previous relay selection with the purpose of providing the maximum coverageextension toward the destination. During the RACR design, we answer to thequestion, where the relay node should be positioned with respect to the physicalenvironment? and, how the coverage extension helps a transmission to be energyefficient? Our performance evaluation shows that RACR achieves a nearly halfreduction in the average path length in dense sensor networks compared to thenon-cooperative geographic routing.

5.1 Introduction

In wireless sensor networks, developing efficient and scalable routing protocolsis one of the challenging tasks that require significant study due to their inherentnature such as the infrastructure absence and the high dynamics (see chapter 2

91

92 Chapter 5. RACR: Relay-Aware Cooperative Routing

for details). Geographic routing, relying on the knowledge of geographic locationinformation of nodes to make local route decisions, emerged as one of the mostsuitable routing solutions in this context. A large number of geographic routingprotocols have been proposed according to different concerns, such as energyconsumption, delay time, overhead expense, quality of service (QoS), and networklifetime, etc., and yet these protocols are based on the premise of using directpoint-to-point communication (direct communication strategy) at the physicallayer.

In the previous chapter, we emphasized the spatial diversity property derivedfrom node cooperation [Laneman 04, Ochiai 05, Sadek 07, Zhong 09, Wang 09a,Vardhe 10] at the physical layer and the lack of interaction with the network andMAC layers.

In addition to this spatial diversity gain from cooperative communication,there is also a potential advantage in terms of coverage extension, which is as-sociated with the link layer connectivity and, hence, affects the routing designand performance at the network layer. However, this is ignored by most existingstudies. There is very little insight into the impact of physical-layer cooperationon the network-layer routing design and performance.

Most of the existing cooperative routing solutions are realized either by findinga route first in a traditional manner and then use cooperative transmission toenhance the link quality along the established route or by building the minimum-power route that applies a cooperation scheme to save the required transmissionpower. Nevertheless, these cooperative routing solutions do not take advantageof coverage extension of physical-layer node cooperation, since the cooperativeroute with coverage extension may be radically different from previous ones.

In considering the demands for high efficiency and scalability in multihop rout-ing as well as the potential benefit of coverage extension from cooperative trans-mission, we investigate how the traditional non-cooperative geographic routing,say Greedy Forwarding (GF), can be improved by incorporating the coopera-tive transmission into the design. In particular, we define the radio coveragebased on the SER requirement to identify the coverage extension due to coop-eration. Furthermore, we address the cooperative geographic routing problemfor SER-constrained multihop sensor networks. The proposed routing protocol,named Relay-Aware Cooperative Routing (RACR), enables a node to make alocal route decision depending on the geographic location of a relay while thisrelay is selected with the purpose of providing the maximum coverage extensiontoward the destination. Simulation results show that in comparison with the non-cooperative GF protocol, the proposed RACR protocol improves significantly therouting performance in terms of the average path length and thus, its stretchfactor.

5.2. System Model 93

Figure 5.1: Extended coverage using cooperative transmission

5.2 System Model

In this section, we describe the signal, channel, and network models for ourcooperation-based routing scheme. Then, we formulate the cooperative greedyrouting problem.

5.2.1 Signal Model

In the following, we introduce the signaling strategies under the direct andcooperative transmission. To do this, Fig. 5.1 depicts an example of how coop-erative transmission leads to a radio coverage extension. We assume that eachnetwork node has the same direct radio range. At the first hop, node S cancommunicate with its neighboring nodes R1, R2, and R3 that are within its di-rect radio range; however, none of them can communicate directly with node D,since D is outside their direct coverage. As exploiting cooperative transmission,D can receive independent symbol copies from different locations, creating thus,a spatial diversity that mitigate the fading effect. Therefore, in this scenario Dcan be reached due to the extended coverage using cooperative transmission.

5.2.1.1 Cooperative Transmission

For the cooperative transmission, we consider a modified version of the three-node decode-and-forward (DF) signaling scheme, as proposed in [Laneman 04].The scheme is set up by a sender S, a relay R, and a receiver D (see Fig. 5.1 herethe relay node R is represented as R1), where each node has a single antenna andcan only either transmit or receive in the current time slot. In the first phase, S


broadcasts its symbol x, while both the receiver and relay receive noisy versionsof x. The received symbols at D and R can be respectively modeled as

yS,D =√PC

1 hS,Dx+ nS,D, (5.1)

yS,R =√PC

1 hS,Rx+ nS,R, (5.2)

where PC1 is the transmission power of S under the cooperative mode, hS,D and

hS,R are the channel coefficients from S to D and R respectively, x is the transmit-ted symbol with unit average power, and both nS,D and nS,R are the noise terms.In the second phase, R decodes the received symbol and then decides whether torelay it. We assume that R correctly decodes the received symbol as long as thereceived signal-to-noise ratio (SNR) is greater than a certain threshold. In thiscase, R performs relaying; otherwise, it remains idle. The received symbol at Dcan be written as

yR,D =√PC

2 hR,Dx+ nR,D, (5.3)

where PC2 is the transmission power of R under the cooperative mode, hR,D is the

channel coefficient from R to D, and nR,D is the noise term. At last, D linearlycombines the received symbols from the two different paths, i.e., yS,D and yR,D,and performs the maximum likelihood (ML) detection. To gain the maximumSNR at the symbol combining output, we consider that D applies the maximumratio combining (MRC) technique [Brennan 03].

5.2.1.2 Direct Transmission

For the direct transmission between S and D, the received symbol at D canbe expressed as

yS,D =√PDhS,Dx+ nS,D, (5.4)

where PD is the transmission power of S under the direct mode.

5.2.2 Channel Model

In this work, large-scale fading and small-scale fading, along with additivenoise, are considered for the channel modeling. Given a transmitter-receiver pair(i, j), we model the channel from i to j as a frequency-flat fading channel withstationary and ergodic time-varying channel coefficient hi,j. It is assumed thatthe channel gain |hi,j| follows a Rayleigh distribution with variance σ2

i,j, that is,hi,j ∼ CN(0, σ2

i,j). Moreover, the channel gain variance is modeled using a σ2i,j ∝

d−αi,j log-distance path loss model [Sadek 07, Li 06, Wang 09b], where α is the path

5.2. System Model 95

loss exponent and di,j is the distance between i and j. We also assume that hi,j isindependent of the channel input and is constant over a symbol duration, and hi,jmay change from a symbol to another as an i.i.d. random process. Regarding theadditive noise term ni,j, we assume that ni,j is a circularly symmetric zero-meancomplex Gaussian random variable with varianceN0, written as ni,j ∼ CN(0, N0).

5.2.3 Theoretical Average SER Performances

In the following we give the theoretical average SER performances for thedirect and cooperative transmission schemes as described above with the widelyused M -QAM modulation.

5.2.3.1 Average SER under Direct Transmission

As shown in [Simon 98], for direct transmission the average SER performanceof M -QAM over frequency-flat Rayleigh fading channels can be obtained by thefollowing closed-form result:

PDSER(γ) = 2K(1− g(γ)) +K2

(4

πg(γ) tan−1

(g−1(γ)

)− 1

), (5.5)

where K = 1− 1√M

, g(γ) =(

1 + 2(M−1)3γ

)−2

, and γ = PDσ2

N0is the average received

SNR per symbol.

5.2.3.2 Average SER under DF Cooperative Transmission

The average SER performance under the DF signaling strategy with M -QAMmodulation over frequency-flat Rayleigh fading channels has been analyzed in[Sadek 08], which provides both the exact and approximate average SER expres-sions as follows. The exact close-form SER is given by

PCSER(γS,D, γS,R, γR,D) = F

(1 +

bγS,Dsin2 θ

)F(

1 +bγS,Rsin2 θ

)+ F

((1 +

bγS,Dsin2 θ

)(1 +

bγR,Dsin2 θ

))(1− F

(1 +

bγS,Rsin2 θ

)),

(5.6)

where F(x(θ)

)= 4K

π

∫ π/20

x−1(θ)dθ − 4K2

π

∫ π/40

x−1(θ) dθ, b = 3/2(M − 1), andγS,D = PC

1 σ2S,D/N0 represents the average received SNR per symbol at node D

from node S; similarly, γS,R = PC1 σ

2S,R/N0 and γR,D = PC

2 σ2R,D/N0. Furthermore,

for sufficiently high SNR, (5.6) can be tightly approximated as


Figure 5.2: Multihop sensor network with cooperative geographic routing.

PCSER(γS,D, γS,R, γR,D) ≈

1

b2γS,D

(A2

γS,R+

B

γR,D

), (5.7)

where

A =M − 1

2M+

(1− 1/√M)2

π,B =

3(M − 1)

8M+

(1− 1/√M)

2

π. (5.8)

5.2.4 Network Model

During this chapter, we consider a multihop wireless sensor network, modeledas a 2-dimensional graph G = (V,E), where V is the set of randomly distributedhomogeneous nodes, with |V | = N , and E is the set of communication linksbetween nodes (i, j) for i, j ∈ V . For any two nodes i, j ∈ V , an edge existsif the transmission from i is received by j with an SNR greater than a requiredthreshold. We consider there is a single session in the network, where data deliverymay cross over multiple hops. Thus, in the node set V , we have two subsets ofnodes: (i) the set of a source-destination pair, Vsd = {s, d}, with |Vsd| = 2, (ii) theset of available nodes, Va = {a1, a2, . . . , aNa} with |Va| = Na, some of them mayserve as intermediate nodes. For clarity, we specify the roles of an intermediatenode in the network as follows. An intermediate node is an active one that helpsdeliver data toward destination. More specifically, based on different purposes ofdata delivering, an intermediate node has two major roles—either a relay nodeor a forwarding node—we call an intermediate node a relay node (e.g., node a in

5.3. SER-Based Radio Coverage Formulation 97

Fig. 5.2) if it is used for cooperative relaying; otherwise, it is called a forwardingnode or forwarder (e.g., node b in Fig. 5.2) if used in the traditional multihopsense.

Fig. 5.2 depicts a cooperative geographic routing scheme for multihop wirelesssensor networks. The multihop routing of interest is realized by a sequence ofcooperative hops that apply the three-node cooperation scheme. As stated above,nodes a, b, and c are the intermediate nodes for the source-destination pair (s, d),where node b serves as a forwarder and both nodes a and c serve as relays. Fig.5.2 also shows that in comparison with the conventional non-cooperative routing,cooperative routing can significantly reduce the number of hops due to the benefitof coverage extension, leading to more efficient communication.

We make the following assumptions for the network model: (i) every nodeknows its own geographic location, (ii) the location of the destination is knownat the source, and (iii) every transmit node uses identical transmission power.The power allocation across the sender and relay as well as the power adjustmentof each node for topology control complicates the cooperative geographic routingproblem. For simplicity, we consider the equal power strategy and PC

1 = PC2 =

PD

2.

5.3 SER-Based Radio Coverage Formulation

In the following we state the radio coverage definition. To identify the coverageextension due to cooperation, we develop a new mathematical expression for radiocoverage, which is somehow different from the previous notion. To do so, we mapthe network graph G = (V,E) into a 2-dimensional geographic plane, where allnodes can be characterized by their geographic location and give the followingdefinitions.

Definition 1: The SER-guaranteed radio coverage R ⊂ R2 of a transmittingside t is defined as a geographic area within which any receive node can meetcertain SER requirement. Formally,

R ={x ∈ R2|PSER(xt,x) 6 ζ0

}, (5.9)

where xt and x denote the geographic locations of the transmitting side andreceiver, respectively, PSER is the average SER at the receiver as a function of lo-cations of both the transmitting side and receiver, and ζ0 is the required averageSER at the receiver. We note that: (i) the transmitting side t can be a singletransmit node using direct transmission or a set of cooperating nodes using coop-erative transmission, (ii) given a communication model, the SER requirement ζ0

translates to a minimum received SNR throughout the radio coverage, (iii) thereare other metrics that could be employed to specify the radio coverage, such as


average bit error rate, outage probability, SNR, etc. These metrics all lead tothis definition; nevertheless, to make the presentation clear we do not try to takeinto account all of these metrics in the formulation but focus on the average SER,and (iv) throughout this chapter we use the term radio coverage to represent thenotion of the SER-guaranteed radio coverage.

From Definition 1, we can define and derive the direct and cooperative radiocoverages as follows.

Definition 2: The direct radio coverage of a transmit node u with transmissionpower PD > 0 is defined as follows:

RD ={x ∈ R2|PD

SER(xu,x) 6 ζ0

}. (5.10)

For direct transmission, the radio coverage contour forms a circle with a radiusrD around the sender because the path loss, modeled as a distance-dependentterm in Section 5.2.2, is the same at a uniform distance from the sender. In Fig.5.2, we show the contour of direct radio coverage based on a fixed transmit powerat the node s.

Definition 3: Based on the log-distance path loss model as described in Sec-tion 5.2.2, the average received SNR can be defined as a distance-dependent term:

γ(di,j) =PDσ2

i,j

N0

=PD

N0dαi,j(5.11)

From Definitions 2 and 3, the direct radio coverage of node u can be writtenin a disk form, i.e.,

RD = D(xu, rD), (5.12)

where xu is the center of the disk and rD is the radius expressed by

rD =

(PD

N0γ0

) 1α

, (5.13)

where γ0 is the required minimum received average SNR to meet the average SERrequirement ζ0. Consider the direct transmission model as introduced in Section5.2.1, given a required average SER ζ0 the corresponding γ0 can be obtained via(5.5). In addition, in this chapter we use the radius rD to specify neighbors withinthe direct coverage for a given node u. Similarly, we will also specify neighborswithin the cooperative coverage by rC .

Definition 4: Given a transmit node u that cooperates with a given set ofrelays Vr = {r1, r2, . . . , rNr}, the cooperative radio coverage of the transmit node

5.3. SER-Based Radio Coverage Formulation 99

u with respect to the relay set Vr is defined as:

RC ={x ∈ R2|PC

SER(xu,xVr ,x) 6 ζ0,xri ∈ RD,

for i = 1, 2, . . . , Nr

}.

(5.14)

Here we note that the cooperating relays are confined to the direct radio coverageof node u. In other words, the cooperating relays have to be one-hop neighborsof node u.

Definition 5: Consider a transmit node u with a set of relays Vr = {r1, r2, . . . , rNr}that can be located at any places within RD, the maximum cooperative radio cov-erage of the transmit node u is defined as:

RCmax =

{x ∈ R2|PC

SER(xu,xVr ,x) 6 ζ0,∀xri ∈ RD,

for i = 1, 2, . . . , Nr

}.

(5.15)

By definitions, it can be seen that RC is a subset of RCmax, and RC

max and is adisk centered at the location of the sender u due to the symmetry of path lossin space. In addition, consider the three-node (s, r, f) cooperation model asintroduced in Section 5.2.1, if we set PC

1 = PC2 = P , since the optimal position of

the relay is on the line segment between the sender and receiver and it approachestoward the middle point of the line segment with increasing the path loss exponent[Wang 09b], it can be shown that

RCmax ≈ D(xs, r

Cmax), (5.16)

where rCmax denotes the radius of the maximum cooperative radio coverage andcan be expressed as

rCmax =

(b2P 2ζ0

N20

(A2kα +B(1− k)α

)) 12α

, (5.17)

where b, A, and B are modulation-dependent constants as given in Section 5.2.3,k is a ratio defined by k , ‖x∗

r−xs‖‖xf−xs‖

, and x∗r is denoted as the optimal relaying

position (on which the relay provides the maximum coverage extension towardthe destination), depending on the path loss exponent α. The calculation of thevalue k refers to [Wang 09b]. Thus, by denoting x∗f as the optimal forwardingposition (on which the forwarder provides the largest progress over the maximumcooperative radio coverage toward the destination), the optimal relaying positioncan be represented as x∗r = k ‖ x∗f −xs ‖ +xs. In Fig. C.7, we give an illustrationof the optimal relaying and forwarding positions as well as the direct and themaximum cooperative transmission radii.


Figure 5.3: RACR Architecture.

5.4 RACR: Relay-Aware Cooperative Routing

In this section, we present the RACR protocol, which is a cross-layer coop-erative routing solution based on a beaconless geographic protocol [Sanchez 09],involving the forwarder and relay selections [Aguilar 10]. The RACR protocoltakes the advantage of coverage extension using the three-node cooperation toselect next hops in a greedy manner. As such, the required number of hops forgiven a source-destination pair can be reduced as compared to non-cooperativegeographic routing. The derived radio coverage formulas for direct and coop-erative transmission schemes as well as the optimal positions for relaying andforwarding are employed in this protocol. Typically, geographic routing proto-cols have two operation modes: greedy and recovery routing modes. However,in this chapter we focus on the design of greedy routing. The recovery routingoperation is not further discussed.

Conceptually, the RACR architecture (see Fig 5.3) is realized through a two-phase selection process. The first phase is to select the best relay such that thesource-relay cooperating pair provides the maximum coverage extension towardthe destination, whereas the second phase is to select the forwarder with thelargest progress toward the destination. Both the relay and forwarder selectionsare based on a distributed contention process without periodic exchange of controlmessages (i.e., beacons) for acquiring neighbors’ location information. During thecontention process, candidate nodes compete for being a relay or forwarder bysetting contention timers relevant to their geographic location.

Given a source-destination pair in the SER-constrained network, the RACRprotocol works as follows. First, the source initiates the two-phase selection pro-cess by broadcasting its message to its direct and cooperative neighbors. Thedirect neighbors decode the message, while the cooperative neighbors hold this

5.4. RACR: Relay-Aware Cooperative Routing 101

Progress toward the destination node

rD

rCmax

x∗fx∗

rxmi

Figure 5.4: Optimal relaying and forwarding positions and the direct and maxi-mum cooperative transmission radii.

message and wait for a relay’s message to perform the maximum ratio combin-ing (MRC). Second, the direct neighbors being able to decode correctly competefor becoming the relay by setting their contention timers as Trelay ∈ [0, Tmax],where Tmax is the maximum delay time allowed for waiting for a relay node. Thecontention timers for relay selection are defined so that candidates located closerto the optimal relaying position x∗r answer first. Third, the winning relay trans-mits its message to its cooperative coverage neighbors, and nodes overhearingthis message cancel their timers. Among the cooperative neighbors, nodes thatcan decode correctly the original message using MRC participate the contention-based forwarder selection process by setting timers as Tfwd ∈ [0, Tmax]. The timersetting for forwarder selection is to let candidates located closer to the optimalforwarding position x∗f answer first. Forth, once the forwarder comes out, itbroadcasts an acknowledgement (ACK) to the source to indicate a correct mes-sage reception, while overhearing nodes cancel their timers. Finally, the forwarderacts as the source node and repeats the same steps until the message reaches thedestination. In the following we describe how we set the contention timers forrelay and forwarder selections, respectively.

5.4.1 Contention Timer Setting for Relay Selection

Since the best relay should be as close to the optimal relaying position x∗r aspossible (see Fig. C.7), each candidate’s timer value must be proportional to thedistance between itself and x∗r. We map this distance into a normalized relayselection metric Mr ∈ [0, 1] as


X

Y

−100 −80 −60 −40 −20 0 20 40 60 80 1000

20

40

60

80

100

120

140

160

180

200

Xri

Xf*

Xr*

Xm

i

Figure 5.5: Example of optimal relay and forwarding positions distributions

Mr =‖ xri − x∗r ‖

rD+ ‖ xs − x∗r ‖, (5.18)

where xri denotes the location of the candidate ri and the denominator rep-resents the farthest distance between a possible candidate’s and optimal relayingpositions. Finally, we set the relay-selection contention timer for each candidateby

Trelay =(Nr − 1)

Nr

Tmax ×Mr + rand

(TmaxNr

), (5.19)

where Nr is the number of groups to be divided in the relaying area andrand(x) is a function obtaining a random value between 0 and x to reduce thecollision probability.

5.4.2 Contention Timer Setting for Forwarder Selection

To select the forwarder with the largest progress over the cooperative ra-dio coverage toward the destination, we set each candidate’s contention timer tobe proportional to the distance between itself and the optimal forwarding posi-tions. To do so, we first define a projection point xmi from the selected relayxri onto the source-destination line, as depicted in Fig. C.7, and θ is given by

θ = arcsin(<xd−xs,xri−xs>‖xd−xs‖‖xri−xs‖

). Given that the coordinates xs, x∗r, and xd are known

by each current node and that each possible relay knows its own xmi position,each candidate forwarder can derive the optimal forwarding position x∗f and set


its corresponding timer Tfwd. We define the forwarder selection metric Mf as

Mf =‖ xfi − x∗f ‖√

(rD)2+ ‖ xmi − x∗f ‖2, (5.20)

where ‖ xmi − x∗f ‖= rCmax− ‖ xs − xmi ‖. Finally, we set the forwarder-selectioncontention timer for each candidate by

Tfwd =(Nf − 1)

Nf

Tmax ×Mf + rand

(TmaxNf

), (5.21)

where Nf is the number of groups to be divided in the forwarding area.

Fig. 5.5 depicts an example of a simple RACR execution. Given given a sourcenode at (0,0), a destination node (0, 300) for rCmax = 180m and rD = 110m. Theoptimal relay position is calculated, the nearest relay node Xri selected at (30,80), then the optimal forwarding position Xf∗ is computed.

5.5 Performance Evaluation

In this section, we effectuate the RACR performance evaluation, to do so, wefirst started by responding to a question related to the relay selection. Shouldwe choose the relay as far as possible from the source node? The answer to thisquestion give us some elements to understand the relationship between the directcoverage definition, the path loss environment and the modulation scheme whendefining the optimum relay location. Then, we analyzed the coverage extensionbenefits using a basic cooperation system, and finally, we determined how thecoverage extension benefits the RACR routing performance in term of energyefficiency.

5.5.1 Should we choose the relay as far as possible fromthe source?

To answer this question, we realized some simulations and found that it isalmost true that the relay node should be as far as possible from the source, butthe position is influenced by the path loss exponent and the modulation typeused by the system. We noted that even if the relay selection tries to get closerto the destination node, RACR establishes that, to maintains a sensor networkwith a SER guaranteed, the relay should be located inside the direct coveragearea. Thus, this important information should be taken into consideration when


Figure 5.6: Relay selection as function of alpha (a)Alpha: 2 (b)Alpha: 3 c()Alpha:4 (d)Alpha: 3.8

defining the optimal relay position at the first phase of the RACR protocol. Fig.5.6 depicts how the optimal relay position may change as function of the path lossexponent. We realized some simulation where we assume the SER requirementζ0 is 10−2, the total transmit power for both the direct and cooperative schemesis P = 15 dBm, the average noise power is N0 = −70 dBm, the modulation typefixed to 16-QAM, and, we varied the path loss exponent α = {2, 3, 3.89, 4}. Fromthis simulations, we observed that in order to maintains the SER requirement,the relay candidates should to be inside rD. If in any case, the optimum relayposition is located outside the rD disk, the point should be displaced to thenearest point within the direct coverage area. The simulations showed us thatthe critical point, that is, the location that provides a better progress towardsthe destination is found when alpha corresponds to 3.89.

5.5.2 Coverage Extension

Then we give the theoretical results to examine the coverage extension fromthe basic three-node cooperation model using a modified decode-and-forwardstrategy over Rayleigh fading channels. We assume the SER requirement ζ0


(a) (b)

(c) (d)

Figure 5.7: Coverage extension (%) with alpha:4, due to cooperation versus the re-laying position with a cross-sectional view for (a) 4QAM, (b) 16QAM, (c)32QAM,(d)64QAM.

is 10−2, the total transmit power for both the direct and cooperative schemes isP = 15 dBm, the average noise power is N0 = −70 dBm, the path loss exponentis α = 4, and the modulation type from 4-QAM to 64-QAM. We compare thecooperative scheme with the direct scheme in terms of the largest progress towardthe destination at (0,∞). For a fair comparison, we set PC

1 = PC2 = PD

2, i.e.,

the total transmit power of the source and relay in the cooperative scheme is thesame as used in the direct scheme. Fig. 5.7 and C.8 depict the increase of radiocoverage as a function of the relay node placed between the source (0, 0) and theforwarder (0, 1) with a normalized distance. It is showed that the best relayingposition is found at the midpoint between S and D, providing about 80% and90% coverage extension toward the destination.


(a) (b)

(c) (d)

Figure 5.8: Coverage extension (%) with alpha:4, due to cooperation versus the re-laying position with a 3D view for (a)4QAM, (b)16QAM, (c)32QAM, (d)64QAM.

5.5.3 Routing performance

Next we give the numerical results to evaluate the routing performance ofthe RACR protocol. The simulation setting are given in Table C.2. We gen-erate 100 network topologies where the nodes are randomly deployed within a1000×1000m2 field. For each topology, we randomly select 750 source-destinationpairs. The simulation results are averaged over the 100×750 simulation runs. Todemonstrate how the proposed RACR protocol improves the non-cooperative ge-ographic routing, we consider the energy efficiency by measuring the number ofhops for the RACR and non-cooperative geographic greedy routing (or namedgreedy forwarding, termed as GF). In the RACR protocol, each hop is based onthe three-node cooperation scheme. In non-cooperative greedy routing, each hopuses direct transmission scheme. As set in the previous simulation, we restricteach hop, whatever it is direct or cooperative, has the same total transmit power.


(a) (b)

Figure 5.9: (a)Average path length versus the average number of neighbors.(b)The corresponding stretch factor.

Thus, the performance metric of the path length (i.e., the number of hops) trans-lates to energy efficiency. Fig. C.9(a) shows the performance of the average pathlength versus the average number of neighbors. It can be seen that the RACRprotocol outperforms significantly the GF protocol due to the effect of coverageextension using cooperation. As considering the corresponding stretch factor,which is defined as the path length ratio, Fig. C.9(b) represents that the RACRprotocol achieves a nearly 50% reduction in average path length compared to theGF protocol. It demonstrates that the proposed RACR protocol is much moreenergy-efficient than the traditional non-cooperative geographic routing protocol.

Table 5.1: Simulation Settings


No. of nodes 2000-2450 Tx. power 15 dBm

Path loss exp. 4 Average noise power -70 dBm

Modulation type QAM Noise figure 15 dBm

Required SER 10e-2 No. of topologies 100

Constellation size 4 No. of simulation runs 75000


5.6 Conclusions

In this chapter, we addressed the cooperative geographic routing problem forSER-constrained multihop sensor networks to look into the impact of physical-layer cooperation on the traditional geographic routing. For given a dense sensornetwork, with the knowledge of geographic location of nodes, the proposed RACRprotocol selects the best relay that provides the maximum coverage extensiontoward the destination, and then selects the next hop in a greedy sense, dependingon the cooperating relay’s location. It is shown that the average path lengthreduction of up to nearly 50% is achievable in dense sensor networks with anSER requirement of 10−2 at the destination.

Chapter 6Conclusion and FutureDirections

The main contribution of this thesis was the design and evaluation of a cross-layer, energy-efficient, local, routing framework able to cope with the wirelesschannel variations. To achieve this global contribution, we studied the BeaconlessGeographic Routing protocols that build the route not only in a local manner,but also on the fly, and, the cooperative communications to exploit the broadcastnature of the wireless channel for Wireless Sensor Networks. During the designand evaluation of our contributions, we considered a scenario of a sensor networkwith SER-level constraints to be satisfied.

We started our research work by realizing an extensive survey of the currentstate-of-the-art of routing and cooperative communications approaches, how theywork, their properties, assumptions and the models used. Subsequently, we stud-ied the requirements of our network scenario, to understand how we could tacklethem by means of a cross-layer system.

Afterwards, our initial intuition was to establish a starting point defined bya beaconless geographic routing strategy that later was used to add our contri-butions as complementary modules, and, build a cross-layer platform. Duringthe development or our study, we identified the cooperative communications as aattractive and suitable mechanism to solve the isolated design in traditional lay-ered sensor networks. Therefore, the result obtained is a cross-layer frameworkconsisting of two routing modules named CoopGeo and RACR. Both protocolshave their own properties and objectives to fulfill.

In the fist part of the thesis, we were interested in reliability of wireless sensornetworks, in effect, many studies have shown that this kind of networks are vul-

109

110 Chapter 6. Conclusion and Future Directions

nerable to network and environment dynamics. This context may cause a loss inthe delivery of packages and the poor use of network resources. Hence, in chapter4, we proposed CoopGeo, a solution that allows an enhancement in packet deliv-ery and the system performance, by means of a cross-layer framework. CoopGeo,consists of two joint cross-layer phases, a joint network-MAC phase for next hopselection and a joint MAC-physical phase for relay selection. In particular, boththe routing and relay selection solutions in CoopGeo are based on geographicinformations using contention-based selection processes.

We have demonstrated by simulation results that our contribution CoopGeocan work with different densities and achieve better system performances thanthe existing traditional geographic routing protocol, in terms of packet error rate,transmission error probability, and saturated throughput.

In the second part of the thesis, we were interested how cooperative commu-nications using its coverage extension property can helps a sensor network to beenergy-efficient. In chapter 5, we proposed RACR protocol that takes the advan-tage of coverage extension using the three-node cooperation to select next hopsin a greedy manner. RACR is realized through a two-phase selection process.The first phase is to select the best relay such that the source-relay cooperatingpair provides the maximum coverage extension toward the destination, whereasthe second phase is to select the forwarder with the largest progress toward thedestination. Both the relay and forwarder selections are based on a distributedcontention process without periodic exchange of control messages (i.e., beacons)for acquiring neighbors’ location information.

We demonstrated by simulations that the required number of hops for givena source-destination pair can be reduced by almost a half in dense wireless sensornetworks.

The results obtained from extensive evaluations of CoopGeo and RACR con-tributions, have demonstrated that both solutions are applicable to sensor net-works in presence very variable channel environments. Therefore, we have provedthat our cross-layer vision of the problem provided an integrated solution to prob-lems like inefficient routing paths, congested medium access and lossy links.

Thus, given the satisfying results of this thesis, we believe the cross-layervision in wireless sensor networks using geographic routing protocols and cooper-ative communications proves to be a practical approach to tackle some problemspresented in these networks.

111

6.0.1 Future directions

We are aware that this theses establish a platform where some other worksmay arise to extend or optimize it such as:

It would be interesting to build a physical testbed where our contributionswere implemented and collect experimental data in order to have a more accurateperformance evaluation. It would be interesting to adapt our contribution to ascenario where the sensor nodes were not assisted by GPS devices, in this context,a virtual coordinates system must be necessary. Although several research worksabout virtual coordinates have been presented in recent years, most of themattack the problem from the network layer viewpoint. It would be interestingpropose a new concept using virtual coordinates that takes into consideration thewireless environment in a beaconless way or at least in a localized way.

We may also think to extend our contributions to include inexpensive andsimple directional antennas that can be used to improve the network communi-cation between the nodes of the network. Directional antennas when radiatingthe signal toward the receiver node may lead to a more efficient utilization of thepower, concentrate the signal diffusion a in better link quality and also increasethe transmission range.

In general, we can say that this thesis paves the way to further protocolsand applications not only in the sensors fields but also in other kind of ad hocnetworks such that vehicular networks.

112 Chapter 6. Conclusion and Future Directions

113

Publications

• S. Syue, V. Gauthier, T. Aguilar, M. Beserra, C. Wang, and H. Afifi, “Relay-Aware Cooperative Routing in Multihop Wireless Networks : A Cross-LayerApproach,” IEEE International Conference on Communications ICC2011,2011 (Submitted).

• T. Aguilar, S. Syue, V. Gauthier, and H. Afifi, W. Chin-Liang “CoopGeo: A Beaconless Geographic Cross-Layer Protocol for Cooperative WirelessAd Hoc Networks,” IEEE Transactions on Wireless Communications (Ac-cepted for major revision).

• T. Aguilar, M.C. Ghedira, S. Syue, V. Gauthier, H. Afifi, and C. Wang,“A Cross-Layer Design Based on Geographic Information for CooperativeWireless Networks,” VTC Spring 20010 - IEEE 71th Vehicular TechnologyConference, 2010, pp. 2-6.

• T. Aguilar and H. Afifi, “Two-hops clustering algorithm with a composedmetric for wireless sensor networks,” The 11th International Symposium onWireless Personal Multimedia Communications (WPMC’08), 2008, pp. 1-5.

• T. Aguilar, H. Afifi, “Analyse de la consommation d’energie des reseauxsans fil IEEE 802.11 au niveau MAC ” Evry : Institut national destelecommunications, 2005 (91-Evry : Impr. INT). - 1 vol. (23 f.) (Collec-tion des rapports de recherche de l’INT , ISSN 0183-0570; n 05006-RS2M).

• T. Aguilar, Hossam Afifi; “Amelioration du mecanisme d’economie d’energiePSM dans le standard IEEE 802.11,” JDIR 2005: 7emes journees Doctor-ales Informatique et Reseau, Troyes, France; Decembre 2005

Annexe AResume du manuscrit dethese en francais

A.1 Vers un protocole de routage geographique

avec contention et communications cooperatives

pour les reseaux de capteurs

Le routage dans les reseaux de capteurs est un service essentiel qui trans-met les lectures des capteurs a certains points de collecte de donnees dans lereseau sur la base des relais multi-saut. Cette tache est particulierement difficilecar elle doit etre realise d’une maniere efficace au niveau de la consommation deresources et avec une quantite limitee d’informations disponibles. La facilite demise a l’echelle et l’utilisation d’informations locales pour fonctionner ont permisau routage geographique d’etre considere comme une approche prometteuse. Ce-pendant, lors de son implementation, certains problemes subsistent en raison desdifficultes pratiques.

Dans ce travail de recherche, deux problematiques inherentes aux protocoles deroutage geographiques ont ete etudiees : i) Le cout associe : aux evanouissementslies aux obstacles et aux multi-trajets suivis par un signal transmis sur un canalradio, aux changements rapides des conditions physiques du canal de transmis-sion et ii) l’administration de resources affectees a chaque noeud appartenant aureseau. Afin de resoudre ces problemes, deux protocoles ont ete presentes : unprotocole de routage geographique avec communications cooperatives, beaconlessCooperative Geographic cross-layer protocol for ad hoc and sensor networks (Co-opGeo) et un protocole de routage base sur le principe d’extension de couvertureRelay-Aware Cooperative Routing (RACR).

115

116 Annexe A. Resume du manuscrit de these en francais

Contrairement aux protocoles de routage geographiques traditionnelles, Co-opGeo est un protocole de routage “beaconless” base sur une architecture inter-couches ou le routage est realise non seulement localement, mais aussi a la vole. Deplus, les problemes lies a la couche physique sont traites par les communicationscooperatives qui exploitent la nature de la diffusion sans fil.

Le protocole RACR exploite la propriete offerte par les communications cooperatives :l’extension de la couverture radio. Cette propriete permet d’ameliorer les perfor-mances d’un reseau que utilise, a l’origine, un protocole de routage geographiquetraditionnel. RACR est une alternative aux scenarios dont l’objectif principal estde diminuer au maximum la consommation des resources du reseau et en memetemps d’assurer que le reseau offre un taux d’erreur par symbole garanti (SER).Ainsi le protocole RACR permet a un noeud d’effectuer des decisions dites locales,par rapport au routage des paquets qui dependent de la localisation geographiqued’un noeud relai, tandis que, ce noeud relai a la finalite de donner une extensionmaximale au niveau de couverture radio en direction de la destination.

Les resultats obtenus a partir des evaluations approfondies de CoopGeo etRACR ont demontre que les deux solutions sont applicables aux reseaux de cap-teurs en presence de forte mobilite, aux environnements tres variables au ni-veau radio, ou avec des erreurs aux niveau de l’information de localisation. Parconsequent, nous avons prouve que notre vision inter-couches du probleme a fournideux solutions efficaces, en termes de chemins, acces au media, problemes lies al’information imprecise de localisation, et des liens perturbes.

Annexe BLa Problematique

Ces dernieres decennies, les technologies sans fil ont connu une croissance ra-pide. Les progres des composants materiels ont suivi la meme tendance permet-tant ansi la production massive de dispositifs de communication comme les ordi-nateurs portables, telephones cellulaires, assistants numeriques personnels (PDA),capteurs, processeurs, etc. Etant donne que ces appareils deviennent plus petitset moins chers et leur association avec certaines technologies conduit a la mise aupoint de nouveaux types de reseaux sans fil, ou de l’amelioration des existants,telles que les reseaux cellulaires (2G, 2.5G, 3G). Un autre type de reseaux sans filse demarque des reseaux sans fil des donnees traditionnels : les reseaux sans filssans infrastructure comme les reseaux ad-hoc, de capteurs et de reseaux mailles.Ces nouveaux types de reseau supportent des nouvelles applications comme leshotspots, les communications en temps reel, reseaux domestiques, les systemesde surveillance, le controle industriel, les reseaux de vehiculaires et les reseaux decapteurs. Les reseaux de donnees sans fil (en faisant reference a la norme IEEE802.11 et derivations) ont ete le centre d’interets scientifiques et commercialespendant plusieurs annees.

Recemment, la communaute scientifique s’est orientee vers les reseaux sans filqui communiquent sans l’aide d’une infrastructure, tel que les reseaux ad hoc et decapteurs. Ce type de reseaux presentent des defis importants dans la conceptionde leur architecture. Les reseaux ad-hoc et de capteurs sont des systemes auto-organisables, formes par des noeuds qui essaient de communiquer les uns avec lesautres.

Les reseaux de capteurs (Recap) appartiennent a la classe des reseaux ad hoc,mais ils ont des caracteristiques supplementaires qui en font un cas particulier.Meme si les reseaux de capteurs partagent des comportements et caracteristiquesavec les reseaux ad hoc, ils ont des differences importantes, par exemple, en raison

117

118 Annexe B. La Problematique

de la petite taille des noeuds qui composent un reseau de capteurs, les noeuds ontdes ressources tres limitees telles que le traitement de l’information, la vitesse duprocesseur, la quantite de memoire, la quantite d’energie disponible et la puissancede transmission. Comme les noeuds sont de petite taille et de faible cout, on peutles deployer en tres grandes quantites, presentant des densites superieures a cellesdes reseaux ad hoc. Nous pouvons egalement ajouter qu’ apres le deploiementdes noeuds, ils restent sans surveillance pendant la duree de vie du reseau (sansentretien ou depannage)

Ainsi les Recap sont plutot concus pour detecter des evenements ou desphenomenes naturels, les noeuds recueillent des lectures, les traitent et les trans-mettent aux utilisateurs. Par la nature meme du reseau, les defis les plus difficilesdans les reseaux de capteurs sont l’efficace gestion de ressources des noeuds et l’adaptabilite aux changements de topologie suivis de l’etat du canal de communi-cation.

B.1 Contexte et defis

Notre travail est oriente vers les reseaux de capteurs sans fil, meme si certainsprotocoles, algorithmes et techniques utilises peuvent etre appliques a d’autrestypes des reseaux. Ainsi, afin de bien definir la portee du travail, nous presentonsles caracteristiques de base et les defis attaches a notre sujet d’etude.

Les reseaux de capteurs sans fil sont composes d’un grand nombre de capteursde petite taille qui sont en mesure de detecter, traiter et communiquer les unsavec les autres. Leur principale finalite est celle de detecter des evenements oudes phenomenes, recueillir et traiter de donnees correspondant aux evenementset les transmettre aux utilisateurs. En plus des caracteristiques liees a la commu-nication sans fil, les reseaux de capteurs presentent d’autres caracteristiques debase heritees de la maniere dont ils travaillent :

• Capacite d’auto-organisation• Communication a courte distance et de routage multi-sauts• Deploiement a haute densite• Changement frequent de la topologie en raison de l’evanouissement du si-

gnal, la mobilite des noeuds et la defaillances de noeuds• Contraintes liees a l’energie, la puissance de transmission, la memoire et la

puissance de calcul• Modele de communication tous vers un

Pendant la conception de l’architecture d’un reseau de capteurs, certainestaches importantes doivent etre considerees, comme, le controle de la consomma-

B.2. Contributions 119

tion de ressources depensees pendant la transmission et la reception de paquets, etla limitation des effets du canal de communications generes par deux phenomenesphysiques : 1) La propagation multi-trajets des ondes electromagnetiques quigenerent des variations dans le signal recu, 2) De l’influence eventuelle de lamobilite des noeuds qui produisent aussi une variation sur le canal de communi-cations. Ainsi, une conception rigoureuse au niveau protocolaire s’impose dans lebut de minimiser la consommation des ressources (i.e. energie) et de maximiserainsi la duree de vie du reseau avec les contrainte liees aux changements de topo-logie de facon a maintenir la connectivite et de calculer les bonnes routes entreles noeuds sources et les noeuds destination.

Afin de reveler ces defis, la communaute scientifique a beaucoup travaille avecdes approches traditionnelles basees sur un modele de couches separees, ainsi,chaque couche de la pile protocolaire (ie. reseau, control d’acces et physique)n’est pas au courant de l’operation de l’autre couche, eliminant ainsi, le beneficede l’optimisation conjointe de l’ensemble des couches qui peut ameliorer les per-formances du reseau.

En consequence, l’utilisation d’une approche inter-couches est obligatoire,celle-ci, doit permettre l’echange de certaines informations importantes entre lescouches, afin de permettre a un noeud de rendre ses decisions de routage plusefficaces, car il aura une vision plus large du comportement du reseau. Cela en-traınera une amelioration de la performance du reseau au niveau global.

Cette etude adopte une approche inter-couches pour ameliorer l’approche aplusieurs niveaux et propose une architecture concue pour faire interagir la couchereseau dans l’acheminement de paquets avec la couche du control d’acces pour ob-tenir l’acces au canal de communications et avec la couche physique afin d’adapterle protocole aux conditions de l’environnement sans fil.

B.2 Contributions

Au cours de ce travail, nous considerons que le routage geographique est unesolution concrete au probleme de routage car la route vers la destination est creea la vole avec des informations locales. Dans ce cadre, l’objectif principal de cettethese est de remplir l’ecart entre les protocoles de routage geographiques tradi-tionnels et l’environnement physique ou les capteurs sont situes. Pour atteindrecet objectif, l’approche utilisee dans la resolution du probleme de routage avecdes liens intermittents se base sur les protocoles de routage geographiques sansbalises ; et aussi l’approche des communications cooperatives qui exploite la na-ture de diffusion des communications sans fils. Ainsi, nos principales contributionssont :

120 Annexe B. La Problematique

• Une conception multi-couches appelee CoopGeo (Cooperative Communi-cations and Geographic Routing) est propose. CoopGeo a ete largementevaluee et comparee a un protocole de routage traditionnel sans balise. Co-opGeo effectue le mecanisme routage geographique base sur la methode decontention et les communications cooperatives avec un mecanisme de selec-tion du relais unique dans le cas ou la communication directe echoue. AvecCoopGeo, nous ameliorons les performance au niveau de la couche physiqueen termes de fiabilite.• Le protocole RACR exploite l’extension de la couverture grace a la coope-

ration des noeuds, afin d’ameliorer le routage geographique non-cooperatifen terme d’efficacite energetique. Il s’agit d’une alternative aux cas ou lesressources du reseau telle que l’energie doit etre preservee tout en respectantune contrainte au niveau de taux d’erreur symbole (SER).

Les resultats obtenus suite a l’evaluation de CoopGeo et RACR nous faitpenser que les deux solutions sont applicables aux reseaux de capteurs en presenceforte mobilite ou dans des environnements avec un canal de transmission tresvariable. Ainsi, nous pouvons dire que notre approche inter-couches du problemepeut fournir une solution integrale aux problemes de routage et au problemesphysiques suivis par les reseaux de capteurs.

Annexe CContributions

C.1 CoopGeo : A Cooperative Geographic Rou-

ting Protocol

L’objet central de notre premiere contribution, est celui d’adapter les pro-tocoles de routage geographiques avec contention au aleas du canal de commu-nications, donc, nous proposons, un systeme qui rend les communications descapteurs, fiables en termes de livraison de bout en bout, et ainsi, faire face al’evanouissement du signal pendant le processus de communication.

CoopGeo est un protocole de communications inter-couches compose de deuxniveaux des conceptions qui travaillent etroitement. Le premier est une concep-tion Reseau-MAC inter-couche qui selectionne d’abord le prochain saut du che-min a parcourir (routage), et le deuxieme est une conception MAC-PHY pour laselectionne du relais (communication cooperative). Les deux niveaux de concep-tion de CoopGeo, sont bases sur l’information geographique des noeuds, et surune methode de contention local controle par des temporisateurs.

C.1.1 Modele de communication

Pour concevoir la plateforme de communications inter-couche, nous consideronsun reseau de capteurs sans fil avec k noeuds deployes aleatoirement dans une zone,ce reseau represente comme un graphe dynamiqueG(V,E), ou V = {v1, v2, . . . , vk}est un ensemble finit de noeuds et E = {e1, e2, . . . , el} un ensemble fini de liensentre les noeuds. On note un sous-ensemble N(vi) ⊂ V , i = 1, . . . , k, comme levoisinage du noeud vi, defini comme l’ensemble des noeuds dans la portee radiode vi.

121

122 Annexe C. Contributions

(a)

(b)

Figure C.1 – (a) Cooperative multihop sensor network model (b) Direct andcooperative modes for each hop

Fig. C.1(a) represente la modelisation du reseau de capteurs, dans laquelle,la source S envoie ses donnees a la destination D de maniere multi-sauts. Danscette figure, le cercle pointille centre sur S illustre sa portee radio. Au debut dechaque transmission, S diffuse ses donnees a ses voisins N(S). Un de ses voisins estchoisi comme le prochain saut (noeud F1) grace a un processus de selection. Deuxmodes de transmission, le direct et le cooperatif, sont consideres a chaque hop. Lemode cooperatif ne fonctionne que lorsque F1 ne peut pas decoder correctementles donnees de S. Apres avoir une version correcte du paquet, F1 agit commele noeud source et repete la meme procedure, et ainsi de suite jusqu’a ce que lepaquet arrive a la destination D.

La Fig. C.1(b) illustre les schemas de transmission des modes directs etcooperatifs, la seule difference entre eux est que F recoit egalement les donneesen provenance de R dans le mode cooperatif, mais pas dans le mode direct. Dans

C.1. CoopGeo : A Cooperative Geographic Routing Protocol 123

la suite nous presentons les modeles des signaux pour les modes de transmissiondirecte et cooperatifs.

Dans le mode direct, S diffuse un symbole x avec une puissance de transmis-sion P , ou la puissance moyenne de x est normalisee a l’unite. Les signaux recuspar F peuvent s’exprimer ainsi :

yS,F =√PhS,Fx+ nS,F , (C.1)

ou hS,F est le coefficient d’evanouissement du canal de S a F , modelise commehS,F ∼ CN(0, σ2

S,F ) ; nS,F est le bruit additif. Pour le mode cooperatif, on appliqueune strategie de selection du relais unique a decodage en deux phases decode-and-forward (DF). Dans la premiere phase, S diffuse son symbole x avec une puissancede transmission Px, tandis que, le prochain saut F et un relais R ecoutent latransmission. Les signaux recus par F et R peuvent etre respectivement exprimesen

yS,F =√PxhS,Fx+ nS,F , (C.2)

yS,R =√PxhS,Rx+ nS,R, (C.3)

Dans la deuxieme phase, le noeud relais selectionne decide s’il transmet lesymbole decode au saut suivant. Si le relais est en mesure de decoder le symboletransmis correctement, il le transmet avec une puissance Px au saut suivant, sinon,il reste inactif. On defini un indicateur IR, tel que :

IR =

{1, if R decodes the transmitted symbol correctly,0, otherwise.

(C.4)

Alors, les signaux recus par le saut suivant pendant la deuxieme phase sont ex-primes ainsi

yR,F =√PxIRhR,Fx+ nR,F , (C.5)

Enfin, le prochain saut combine les signaux recus, en utilisant le mecanismede maximum de combinaison (MRC).

yF =√Pxh

∗S,FyS,F +

√PxIRh

∗R,FyR,F . (C.6)

Ainsi, le symbole decode par le saut suivant x est exprime par l’equation

x = arg minx∈A

|yF − Px(|hS,F |2 + IR|hR,F |2)x|2, (C.7)

Ou |A| = Θ denote la cardinalite de la constellation Θ-aire.Selon l’analyse des performances de [Su 08], le taux d’erreur de symboles


Figure C.2 – Area division for CoopGeo routing. F1 and F2 are sub-area 0 and 1of PPA respectively, whereas F3 and F4 are sub-area 4 and 5 of NPA respectively.

(SER) au noeud suivant peut etre exprime comme

Ps ≈4N2

0

b2P 2xσ

2S,F

(A2

σ2S,R

+B

σ2R,F

), (C.8)

qui est une approximation dans un regime de haut SNR. Quand M -QAM estutilise, b = 3/2(M − 1), et

A =M − 1

2M+

(1− 1/√M)2

π,B =

3(M − 1)

8M+

(1− 1/√M)

2

π. (C.9)

C.1.2 CoopGeo : A geographic cross-layer protocol for co-operative wireless networks

Le protocole CoopGeo, en general, effectue deux taches dans un reseau decapteurs cooperatif multisaut : le routage et la selection du relais. Comme decritauparavant, le processus de routage CoopGeo fonctionne en deux phases, a savoir,les phases dites BLGF et BLRF. Dans la phase BLGF, un noeud qui joue lerole de prochain saut dans le routage et qui fournit le progres maximal versla destination est choisi en utilisant un processus de contention base sur destemporisateurs. Quand la phase BLGF echoue dans la recherche du prochainsaut, le processus de routage entre dans la phase BLRF et applique le mecanisme“face routing” en utilisant la planarisation du reseau base sur le principe de“selection et protestation”. La selection du relais (relais cooperatif) s’effectuequand le noeud designe comme prochain saut n’est pas arrive a decoder le paquetcorrectement. Dans ce cas, CoopGeo demarre la tache de selection du relais, afin

C.1. CoopGeo : A Cooperative Geographic Routing Protocol 125

de trouver un noeud relais optimal qui offre le meilleur lien cooperatif entre lenoeud source et le prochain saut.

Fig. C.1(a) donne un exemple de selection au niveau du routage et du relaisdans CoopGeo. Les noeuds localises dans l’aire PPA participent dans la phaseBLGR, a savoir, X1, X2, R1, and F1. Ceux situes dans l’aire NPA, c’est-a-direW1, . . . ,W4, sont consideres comme participant de la phase BLRF.

Le noeud F1 est choisi comme le prochain saut du noeud source S. La trans-mission des donnees entre la source S et de son prochain saut F1 est realisee atravers une transmission directe ou cooperative. Les noeuds relais candidats de lapaire source-prochain saut (S, F1) qui participent dans le mecanisme de selectionde relais, sont ceux au sein de la zone de relayage (qui sera definie plus tard), asavoir R1 et X1. Dans cette figure R1 est selectionne comme relais optimal dansle mode cooperatif.

C.1.3 Transmission greedy sans balises de controle (BLGF)

Au debut d’une transmission, S declenche le processus BLGF. en diffusant unpaquet dans son voisinage, puis il attend la reponse du meilleur saut au cours d’untemps Tmax/2 . Pendant cette periode, le voisinage est en concurrence pour trans-mettre le paquet par l’etablissement des temporisateurs (TCBF ), comme expliquedans la section C.1.3.1. Lorsque le meilleur candidat envoie Clear to Forward(CTF) a S en raison de l’expiration de son temporisateur, les autres candidats,en entendant ce paquet, suppriment leurs temporisateurs, et finalement, la sourceet le noeud gagnant la contention realisent un echange de paquets de controle (SE-LECT / ACK) afin d’indiquer qu’un noeud intermediaire a ete trouve. A partir dece moment, le noeud intermediaire devient la source des donnees et la procedurese repete.

C.1.3.1 Selection a base de temporisateurs (TCBF )

Pour mettre en oeuvre les temporisateurs TCBF des noeuds intermediairescandidats, nous avons applique les temporisateurs en fonction de leur localisation.La figure C.2 represente la couverture radio d’un noeud source, divise en deuxzones : PPA et NPA. Comme mentionne auparavant, celles ci, sont a la fois divisesen sous-regions appelees sous-zones communes (CSAs).

Le reglage des temporisateurs de chaque noeud est donne comme suit. D’abord,chaque noeud situe dans la zone PPA identifie a quel groupe CSA il appartient

CSAPPA =⌊NSA× r − (dS,D − dFi,D)

2r

⌋(C.10)

ou NSA est un nombre predefini sous-domaines qui divisent la zone de cou-


verture, r le progres maximum ou la portee de transmission, et (dS,D − dFi,D)represente les progres d’un noeud intermediaire candidat a la destination (lesnoeuds localises dans NPA utilisent Eq. (C.12) pour obtenir leur CSA). Ensuite,etant donne CSAPPA ou CSANPA, chaque candidat calcule son temporisateuravec l’equation suivante :

TCBF =(CSA× Tmax

NSA

)+ rand

( TmaxNSA

)(C.11)

ou Tmax represente le retard maximum pendant lequel le noeud source S attendune reponse d’un noeud intermediaire et rand(x) est une fonction qui donne unevaleur aleatoire entre 0 et x pour reduire la probabilite de collision des paquets.La fonction TCBF alloue la premiere moitie du Tmax aux candidats situes dansPPA pour la phase BLGF et l’autre moitie aux candidats situes dans NPA pourla phase BLRF.

C.1.3.2 Recuperation greedy sans balises de controle (BLRF)

Cette procedure de recuperation est initiee quand la transmission greedy netrouve pas un noeud intermediaire dans la zone PPA. Ainsi, la recherche d’unnoeud dans la zone NPA est declanchee automatiquement. Pour ce faire, nousavons applique le mecanisme de recuperation propose par [Kalosha 08]

La zone NPA est divisee en n = NSA2

couronnes concentriques de taille egale,

ou la largeur de la couronne ith est (√i −√i− 1)r1, et r1 est le rayon de la

premiere couronne r1 = r√n. Afin d’utiliser la meme terminologie, a partir de

maintenant, nous nous refererons a une couronne comme un groupe CSA. Parconsequence, un noeud v ∈ NPA identifie son CSA de maniere similaire a cellede noeuds dans la zone PPA

CSANPA =⌊(√n · dv,u

r

)2⌋+NSA

2(C.12)

A ce moment, connaissant son CSA, le noeud determine la valeur de sontemporisateur en utilisant la meme equation que les noeuds situes dans la zonePPA (Eq. C.11).

C.1.4 Selection du relais base sur des informations geographiques

Un critere de selection de relais sur la base d’informations geographiques, ou lemeilleur relais est determine en fonction d’une metrique mi est utilise. Le criterede selection de relais pour chaque saut cooperatif peut etre exprime comme suit,

i∗ = arg mini∈{1,2,...,N}

mi = arg mini∈{1,2,...,N}

A2dpS,Ri +BdpRi,F , (C.13)

C.2. Evaluation des Performances 127

ou dS,Ri et dRi,F sont les distances entre le noeud source le i-eme noeud relais,et entre le i-eme noeud relais et le prochain saut intermediaire, respectivement,et A et B sont des constantes de modulation dependant que satisfont l’equation(C.9). Nous notons que le meilleur noeud relais choisi par le critere ci-dessus estcelui qui fournit le meilleur lien source-relais cooperatif en matiere de SER moyendans le prochain saut (le prochain noeud intermediaire).

Le processus de selection du relais commence des que chaque noeud relaisecoute l’echange DATA/CTF. Il demarre son temporisateur si le noeud intermediairedemande au voisinage de l’aide pour arriver a decoder le paquet recu. Une foisque le temporisateur du meilleur relais expire, il envoie immediatement le paquetdemande par le noeud source. Nous avons precedemment defini la metrique pourla selection du relais, celle qui minimise le SER en fonction du schema de modula-tion utilise, ou A et B sont deux constantes dependant du schema de modulation.Le meilleur relais xi, dont la metrique est f(xi), serait alors, le plus proche de x∗

qui satisfait l’equation (C.17) provenant de (C.16).

mi , A2dpS,Ri +BdpRi,F , i = 1, 2, . . . , N, (C.14)

f(xi) = A2 ‖xi − xS‖p +B ‖xi − xF‖p (C.15)

minimise f(xi) = A2 ‖x− xS‖p +B ‖x− xF‖p (C.16)

x∗ =A2xS +BxFA2 +B

(p = 2) (C.17)

Nous obtenons une fonction M, qui mappe, la fonction f dans le intervalle[0, 1], ou xmax est le point dans un ensemble :

M(f(x)) =f(x)− f(x∗)

f(xmax)− f(x∗)(C.18)

Enfin, nous utilisons l’equation suivante pour allouer le temps a chaque noeuddans le schema de selection relais base sur la contention (CBR)

TCBR = Tmax M(f(x)) + rand(2TmaxNSA

)(C.19)

C.2 Evaluation des Performances

Notre methodologie de simulation evalue les performances au niveau PHY/MACavec des simulations Monte-Carlo mises en oeuvre sur Matlab code. Nous avonssimule les trois processus des couches inferieures. Les parametres des simulationssont exprimes dans la table C.1. Les resultats obtenus se basent sur 20,000 topolo-gies generees aleatoirement, ou tous les noeuds sont en concurrence pour acceder


Table C.1 – Simulation Settings


No. of Neighbors 1-20 Tx. Power 25 dBm

Channel Model Rayleigh Average Noise 20 dB

Carrier Frequency 2.412 Ghz Noise Figure 15 dB

Channel Bandwidth 22 Mhz Packet Size 1538 Bytes

Modulation Type QAM No. of Topologies 20000

Constellation Size 4-128 No. of Simu. Trials 2000000

Contention Period 500 µs

au canal de transmission. Dans les simulations, nous commencons par trouver lesnoeuds intermediaires et noeuds relais qui participent a chaque saut vers la des-tination finale, et une fois ces noeuds obtenus, nous les utilisons pour evaluer letaux d’erreur symbole sur les paquets, la probabilite de transmission moyenne, ledebit sature, et d’autres experimentations derivees de la variation des parametres.

C.2.1 Taux d’erreur de paquets (PER)

Afin d’analyser le PER de BOSS et CoopGeo, nous commencons par simulerles protocoles avec des valeurs Tmax differents, allant de Tmax = 100µs jusqu’aTmax = 1000µs. Les resultats sont presentes dans les figures C.3(a) C.3(b).

Les deux figures nous permettent d’avoir une visualisation globale du compor-tement des protocoles a l’egard du PER. A partir de ces resultats, nous pensonsque le comportement lorsque CoopGeo utilise Tmax = 500µs represente un bonnerapport entre le PER et le delai necessaire pour choisir les noeuds cooperatifsd’une transmission. Ainsi, dans la figure C.4, nous montrons que le PER moyende deux protocoles avec cette valeur choisie. Le taux d’erreur sur les paquets dela figure inclut a la fois la probabilite de collision dans les periodes de contentionet la probabilite d’erreur sur le canal sans fil.

Nous prouvons que notre protocole a presente une baisse du taux d’erreur de2,5 fois moins, par rapport au protocole de routage geographique traditionnel.Nous remarquons aussi que le taux d’erreur des protocoles se rapproche l’un del’autre en fonction de la densite de noeuds en raison de la probabilite de collisiondans les periodes de contention.

C.2. Evaluation des Performances 129

(a) PER for BOSS (b) PER for CoopGeo

Figure C.3 – PER de BOSS et CoopGeo avec Tmax = 100...1000 µs. Les courbessituees dans la partie inferieure du graph correspondent a Tmax = 100 et cellessituees dans la partie superieure correspondent a la valeur maximale Tmax = 1000

C.2.2 Probabilite d’erreur dans la transmission de bouten bout

Dans la figure C.5, nous montrons que la probabilite moyenne d’avoir uneerreur dans la transmission est nettement superieur dans le cas cooperatif. Memele taux diminue avec l’augmentation du nombre des noeuds qui sont candidats arelayer le paquet. Ce comportement est du a la selection precise du noeud relaislorsque plusieurs noeuds sont presents dans le voisinage. Nous pouvons egalementremarquer que CooopGeo experimente un taux d’erreur de transmission faible quinous permet d’augmenter la taille de la constellation du schema de modulationafin de profiter au maximum la bande passante de bout en bout.

C.2.3 Variation des parametres d’entree

1) La variation de la fenetre de contention Tmax : Dans cet experimentation,nous etudions l’impact de la taille de la fenetre de contention Tmax (controle leretard affecte a un noeud concurrent quand il essaie de transmettre un paquetcomme intermediaire ou relais) sur la performance de CoopGeo. Nous simulonsle protocole avec des valeurs Tmax de 100µs a 1000µs.

Dans la figure C.6(a), nous constatons que les collisions causees par les noeudsen contention (noeuds intermediaires et relais) diminuent avec l’augmentation dela taille de Tmax. Les tailles de 500µs a 1000µs sont les mieux adaptees pourCoopGeo.


2 4 6 8 10 12 14 16 18 20 220

0.1

0.2

0.3

0.4

0.5

0.6

0.7

Contending nodes

Pac

ket E

rror

Rat

e

BOSSCoopGeo

Figure C.4 – Packet Error Rate for Tmax = 500µs

2) La variation de la taille de constellation de la modulation : la figureC.6(b),d) demontre que CoopGeo a une meilleur performance en terme de debitsature (saturated throughput) quand la taille de la constellation est augmentee.En raison d’un taux d’erreur tres faible dans CoopGeo, nous pouvons augmen-ter la taille de constellation en fonction des environnements de transmission sansdeteriorer le debit sature de bout en bout.

C.3. RACR : Relay-Aware Cooperative Routing 131

2 4 6 8 10 12 14 16 18 20 220

0.05

0.1

0.15

0.2

0.25

Contending nodes

Err

or T

rans

mis

sion

Pro

babi

lity

BOSSCoopGeo

Figure C.5 – End to End Transmission Error Probability for Tmax = 500µs

C.3 RACR : Relay-Aware Cooperative Routing

Notre objectif dans cette section est de concevoir un systeme efficace, ca-pable de controler la consommation des ressources des noeuds qui composent unreseau de capteurs. Pour faire face a cet objectif, nous presentons notre deuxiemecontribution : un protocole de routage geographique cooperatif avec connaissancedu relais, Relais-Aware Cooperative Routing (RACR), qui exploite la proprieted’extension de couverture due a la cooperation des noeuds.

Le protocole RACR permet a un noeud intermediaire (forwarding) qui sup-porte un taux d’erreur symbole (SER) defini a participer dans le processus de rou-tage par le biais d’une decision locale qui se base sur une selection prealable d’unnoeud relais avec le but de fournir une couverture maximale vers la destination.Lors de la conception RACR, nous repondons a la question, “comment l’extensionde couverture permet a une transmission d’etre efficace au niveau energetique ?”L’evaluation des performances montre que RACR reduit de moitie la longueur duchemin moyen pendant les transmission de donnees dans les reseaux de capteursdenses par rapport a un protocole de routage non-cooperatif.

C.4 SER-Based Radio Coverage Formulation

Dans cette section, nous indiquons de nouvelles formulations qui seront uti-lisees par le protocole RACR. Pour identifier l’extension de couverture grace ala cooperation, nous developpons une nouvelle expression mathematique pour lacouverture radio. Pour ce faire, nous avons defini un reseau G = (V,E) dans un


2 4 6 8 10 12 14 16 18 20 220

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1

Contending nodes

Col

lisio

n P

roba

bilit

y

Tmax = 100Tmax = 200Tmax = 300Tmax = 500Tmax = 800Tmax = 1000

(a)

0 2 4 6 8 10 12 14 16 18 203

3.2

3.4

3.6

3.8

4

4.2

4.4

4.6

4.8x 10

6


Dat

a ra

te

Saturation throughput for QAM

DirectCoop 16QAMCoop 32QAMCoop 64QAMCoop 128QAM

(b)

Figure C.6 – (a)CTF-Relayed message collision probability when changing Tmaxfrom 100µs to 1000µs. (b)CoopGeo Saturated throughput for QAM fom 16-128.

C.4. SER-Based Radio Coverage Formulation 133

espace a 2 dimensions, ou les noeuds connaissent leur positions geographiquesavec une strategie allocation de puissance PC

1 = PC2 = PD

2.

Definition 1: La couverture radio avec un SER guarantee R ⊂ R2 d’un noeudemetteur t est defini comme la zone geographique ou un noeud recepteur guaran-tees une demande d’un certain SER. Formalement,

R ={x ∈ R2|PSER(xt,x) 6 ζ0

}, (C.20)

ou xt et x designent les positions du transmetteur et recepteur, respectivement,PSER est le SER moyen au niveau du recepteur en fonction des noeuds sources etrelais, et ζ0 est le SER moyen requis au niveau du recepteur. Nous notons que :(i) La tranmission peut etre directe ou cooperative, (ii) la demande SER ζ0 setraduit par un niveau de SNR, (iii) Dans cette section le terme couverture radioequivaut au terme couverture radio avec un SER guarantee.

A partir de cette definition, nous derivons les definitions de couverture radiodirecte et cooperative.

Definition 2: La couverture radio directe d’un noeud source u avec une puis-sance de transmission PD > 0 est definie :

RD ={x ∈ R2|PD

SER(xu,x) 6 ζ0

}. (C.21)

Definition 3: En utilisant le modele d’evanouissement log-distance, le rapportsignal bruit moyen SNR peut etre defini en terme de la distance :

γ(di,j) =PDσ2

i,j

N0

=PD

N0dαi,j(C.22)

Ansi, la couverture radio directe d’un noeud est representee par un disk,

RD = D(xu, rD), (C.23)

ou xu represente le centre du disque et rD le rayon de couverture, exprime ainsi :

rD =

(PD

N0γ0

) 1α

, (C.24)

ou γ0 represente le SNR moyen requis pour repondre au SER moyen ζ0. Danscette section, nous utilisons le terme rD pour preciser les voisins dans le rayonde couverture directe d’un noeud u, egalement, nous faisons reference au voisinsdans le rayon de couverture cooperative en utilisant le terme rC . De meme, nousallons egalement specifier voisins dans la couverture de cooperation par rC.

Definition 4: Etant donne un noeud u qui coopere avec un ensemble de noeudsrelais Vr = {r1, r2, . . . , rNr}, la couverture radio cooperative du noeud u par rap-


port a l’ensemble de relais Vr est definie ainsi :

RC ={x ∈ R2|PC

SER(xu,xVr ,x) 6 ζ0,xri ∈ RD,

for i = 1, 2, . . . , Nr

}.

(C.25)

Nous notons que les relais sont confines dans la couverture radio directe du noeudu. En d’autre terme, le relais cooperatif doit etre un voisin a un saut de u.

Definition 5: La couverture radio maximale cooperative du noeud u est definie :

RCmax =

{x ∈ R2|PC

SER(xu,xVr ,x) 6 ζ0,∀xri ∈ RD,

for i = 1, 2, . . . , Nr

}.

(C.26)

Par definition, on peut voir que RC est un sous-ensemble de RCmax, et RC

max, estun disque centre a l’emplacement de u. Comme la position optimale du relaisse localise sur le segment de la ligne entre l’emetteur et le recepteur, on peutdemontrer que :

RCmax ≈ D(xs, r

Cmax), (C.27)

ou rCmax denote le rayon de la couverture radio maximale et peut s’exprimer ainsi :

rCmax =

(b2P 2ζ0

N20

(A2kα +B(1− k)α

)) 12α

, (C.28)

ou b, A, et B sont des constantes dependant de la modulation, k correspond ak , ‖x∗

r−xs‖‖xf−xs‖

, et x∗r represente la position optimale du relais (sur laquelle le relais

peut fournit l’extension de couverture maximale a l’egard de la destination),en fonction de l’exposant d’evanouissement. Pour calculer la valeur de k voir[Wang 09b]. Ainsi, avec x∗f qui denote la position optimale du noeud intermediairequi fournit le plus grand progres vers la destination, la position optimale du noeudrelais peut etre exprimee ainsi : x∗r = k ‖ x∗f − xs ‖ +xs. La figure C.7 illustre lespositions optimales du noeud intermediaire et du noeud relais et les couverturesdirectes et cooperatives.

C.5 Architecture de RACR

Dans cette section, nous presentons notre deuxieme contribution. RACR estun protocole cooperatif inter-couches, base sur un protocole de routage geographiquesans balise de controle [Sanchez 09], qui implique la selection des noeuds in-termediaires et relais [Aguilar 10]. RACR utilise le principe d’extension de cou-verture a travers la cooperation de trois noeuds. Etant donne une pair source-

C.5. Architecture de RACR 135

destination, le nombre de saut du chemin peut etre reduit par rapport au pro-tocoles de routage geographiques non-cooperatifs. Pour ce faire, nous avons uti-lise les formules presentees precedemment, telles que, la couverture radio directeet cooperative, ainsi que les positions optimales du relais et intermediaires. Demaniere generale, le routage geographique presente deux modes de fonctionne-ment : le glouton (greedy) et celui de recuperation du chemin. Cependant, RACRest concentre sur la conception du routage glouton.

L’architecture RACR se realise a l’aide de deux processus de selection. Le pre-mier processus est de selectionner les meilleurs relais afin que les noeuds source-relais cooperent afin de fournir une extension de couverture maximale a l’egardde la destination, alors que la deuxieme phase consiste a selectionner le noeudintermediaire (forwarding node) avec les plus grand progres vers la destination.Les deux selections sont fondees sur un processus distribue qui evite l’echangeperiodique de messages de controle (les balises) pendant l’acquisition des infor-mations de localisation des voisins. Au cours des processus de selection, les noeudsen competition (relais ou intermediaire) reglent leurs temporisateurs par rapporta leur situation geographique.

Etant donne une paire source-destination dans un reseau avec des contraintesau niveau SER, RACR fonctionne de la maniere suivante. D’abord, la sourceinitie le processus de selection a deux phases par la diffusion de son message ases voisins directs et cooperatifs. Les voisins directs decodent le message, tandisque les voisins cooperatifs maintiennent ce message et attendent une deuxiemeversion du message d’un relais pour effectuer leur combinaison (MRC). Apres,les voisins directs qui ont decode correctement le message concurrencent pourdevenir le noeud relais en utilisant leur temporisateur Trelay ∈ [0, Tmax], ou Tmaxest le delai maximum autorise d’attente d’un noeud relais. La conception destemporisateurs est telle que les noeuds relais situes le plus pres de la positionoptimale x∗r repondent en premier. Ensuite, le noeud relais choisi transmet lemessage a ses voisins localises dans la couverture cooperative, et les noeuds quientendent ce message annulent leurs temporisateurs. Parmi les voisins cooperatifs,les candidats qui sont capables de decoder correctement le message participentau processus de selection du noeud intermediaire en definissant aussi des tempo-risateurs Tfwd ∈ [0, Tmax]. De la meme maniere que le cas des temporisateurs desnoeuds relais, ces temporisateurs sont regles tels que, les candidats les plus presde la position optimale x∗f repondent en premier. Alors, une fois que le noeudintermediaire est choisi, il diffuse un accuse de reception (ACK) au noeud sourcepour indiquer la reception correcte du message, tandis que les autres noeudsannulent leur temporisateurs quand ils ecoutent cette transmission. Finalement,le noeud intermediaire devient le noeud source et les deux phases de RACR serepetent. Maintenant, nous detaillons le reglage des temporisateurs.


Progress toward the destination node

rD

rCmax

x∗fx∗

rxmi

Figure C.7 – Positions optimales du relais et intermediaire, et rayon de trans-mission direct et cooperatif.

C.5.1 Selection du relais

Comme le meilleur noeud relais doit etre le plus proche de la position optimaledu relais x∗r (voir figure C.7), les temporisateurs des noeuds doivent etre propor-tionnelles a la distance entre le noeud meme et x∗r. Pour ce faire, nous mapponscette distance dans une metrique de selection normalisee Mr ∈ [0, 1], definie

Mr =‖ xri − x∗r ‖

rD+ ‖ xs − x∗r ‖, (C.29)

Ou xri designe l’emplacement du candidat ri et le denominateur represente laplus grande distance entre un candidat et l’emplacement optimal du relais. Enfin,nous avons defini le delai affecte a chaque temporisateur comme

Trelay =(Nr − 1)

Nr

Tmax ×Mr + rand

(TmaxNr

), (C.30)

Ou Nr designe le nombre de groupes qui forment la zone de relais et rand(x)nous donne une valeur entre 0 et x, afin de reduire la probabilite de collision desnoeuds dans le meme groupe.

C.5.2 Selection du noeud intermediaire

Pour selectionner le noeud intermediaire dans un saut (le noeud avec le plusgrand avancement dans la couverture radio cooperative vers la destination), lestemporisateurs de chaque noeud doivent etre proportionnelles a leur distance avecles position optimales de transmission cooperatives. Pour ce faire, nous avonsdefini un point de projection xmi a partir du noeud relais choisi xri vers la ligne

C.6. Evaluation des performances 137

formee par les noeuds source et destination, comme illustre la figure C.7, ou θ

est donne par θ = arcsin(<xd−xs,xri−xs>‖xd−xs‖‖xri−xs‖

). Etant donne que les coordonnees xs,

x∗r, et xd sont connues par le noeud actuel et que chaque noeud relais candidatconnaıt sa propre position xmi , chaque noeud intermediaire qui est candidat estcapable de deriver la position optimale intermediaire x∗f et determiner ainsi sontemporisateur Tfwd. Donc, la metrique Mf de selection d’un noeud intermediaireest

Mf =‖ xfi − x∗f ‖√

(rD)2+ ‖ xmi − x∗f ‖2, (C.31)

ou ‖ xmi − x∗f ‖= rCmax− ‖ xs − xmi ‖. Finalement, pour affecter le delai autemporisateur de chaque noeud, on calcule :

Tfwd =(Nf − 1)

Nf

Tmax ×Mf + rand

(TmaxNf

), (C.32)

ou Nf represente le nombre de groupes qui forment la zone de selection des noeudsintermediaires.

C.6 Evaluation des performances

C.6.1 Extension de la couverture

D’abord, nous analysons les resultats theoriques de l’extension de la couver-ture du modele cooperatif avec un canal d’evanouissement Rayleigh. On sup-pose que le SER demande est ζ0 = 10−2, la puissance de transmission totalepour les deux regimes (directs et de cooperatif) est P = 15 dBm, le bruitmoyen est N0 = −70 dBm, l’exposant d’evanouissement α = 4, et les constel-lation de la modulation a partir de 4-QAM jusqu’a 64-QAM. Nous comparonsles regimes de cooperation avec le regime direct en termes d’avancement versla destination (0,∞). Afin de produire une comparaison equitable, nous avons

defini PC1 = PC

2 = PD

2(la puissance de transmission totale du noeud source et

du noeud relais dans le regime cooperatif est la meme que celle utilisee dans leregime direct. La figure C.8 represente l’extension obtenue de la couverture radioen fonction des noeuds relais places entre la source (0, 0) et le noeud intermediaire(0, 1) (distance normalisee). On note que la meilleure position du relais se trouvea mi-chemin entre S et D. Les simulations montrent une extension de covertured’environ 80% et 90% vers la destination.


(a) (b)

(c) (d)

Figure C.8 – Coverage extension (%) with alpha :4, due to cooperation ver-sus the relaying position with a 3D view for (a)4QAM, (b)16QAM, (c)32QAM,(d)64QAM.

C.6.2 Efficacite energetique

Ensuite nous donnons des resultats numeriques pour evaluer la performanceau niveau routage du protocole RACR. Le reglage des simulations est donne dansle tableauC.2. Dans les simulations, nous generons 100 topologies ou les noeudssont deployes aleatoirement dans un espace de 1000×1000 m2. Pour chaque to-pologie, nous selectionnons de facon aleatoire 750 paires source-destination. Pourdemontrer l’efficacite energetique du protocole RACR par rapport a une approchegeographique non-cooperatif, nous considerons que l’efficacite energetique est me-suree avec le nombre de sauts de la route. Dans notre experimentation, nous avonsconsidere la phase de routage greedy d’un protocole de routage geographique tra-ditionnel (Greedy Forwarding). Dans RACR, chaque saut se base sur le regime

C.6. Evaluation des performances 139

(a) (b)

Figure C.9 – (a)Average path length versus the average number of neighbors.(b)The corresponding stretch factor.

cooperatif a trois noeuds. Dans le routage ”greedy”, chaque saut utilise le regimede transmission directe.

Ainsi, l’indicateur de la performance de RACR correspond a la longueur duchemin (a savoir, le nombre de sauts) qui traduit l’efficacite energetique. La figureC.9(a) montre la longueur moyenne du chemin par rapport au nombre moyen devoisins. Nous pouvons constater que RACR surpasse significativement le protocoleGF en raison de l’extension de la couverture comme produit de la cooperationentre les noeuds. Une autre moyen d’evaluer l’efficacite energetique est de calculerle facteur d’etirement ”stretch factor”, la figure C.9(b) montre une reductiond’environ 50% de la longueur du chemin par rapport a celle du protocole GF.Ainsi, nous considerons que RACR est plus efficace au niveau energetique.

Table C.2 – Simulation Settings


No. of nodes 2000-2450 Tx. power 15 dBm

Path loss exp. 4 Average noise power -70 dBm

Modulation type QAM Noise figure 15 dBm

Required SER 10e-2 No. of topologies 100

Constellation size 4 No. of simulation runs 75000

Bibliography

[Adam 08] Helmut Adam, Christian Bettstetter & Sidi MohammedSenouci. Adaptive relay selection in cooperative wireless net-works. 2008 IEEE 19th International Symposium on Personal,Indoor and Mobile Radio Communications, pages 1–5, septem-bre 2008.

[Adam 09] Helmut Adam, Christian Bettstetter & Sidi MohammedSenouci. Multi-Hop-Aware Cooperative Relaying. VTC Spring2009 - IEEE 69th Vehicular Technology Conference, pages 1–5,avril 2009.

[Aguilar 10] T. Aguilar, M. C. Ghedira, S.-J. Syue, V. Gauthier, H. Afifi &C.-L. Wang. A cross-layer design based on geographic infor-mation for cooperative wireless networks. In Proc. 2010 IEEEVehic. Tech. Conf. (VTC 2010-Spring), Taipei, Taiwan, May2010.

[Ahlswede 00] R. Ahlswede, Ning Cai, S.-Y.R. Li & R.W. Yeung. Networkinformation flow. Information Theory, IEEE Transactions on,vol. 46, no. 4, pages 1204 –1216, jul 2000.

[Akyildiz 02] I.F. Akyildiz, W. Su, Y. Sankarasubramaniam & E. Cayirci.A survey on sensor networks. IEEE Commun. Mag., vol. 40,pages 102–114, 2002.

[Aurenhammer 91] Franz Aurenhammer. Voronoi diagrams—a survey of a fun-damental geometric data structure. ACM Comput. Surv.,vol. 23, no. 3, pages 345–405, 1991.

[Basagni 98] Stefano Basagni, Imrich Chlamtac, Violet R. Syrotiuk &Barry A. Woodward. A distance routing effect algorithm formobility (DREAM). In MobiCom ’98: Proceedings of the 4thannual ACM/IEEE international conference on Mobile com-puting and networking, pages 76–84, 1998.

141

142 Bibliography

[Bettstetter 03] Christian Bettstetter, Giovanni Resta & Paolo Santi. TheNode Distribution of the Random Waypoint Mobility Modelfor Wireless Ad Hoc Networks. IEEE Transactions on MobileComputing, vol. 2, no. 3, pages 257–269, 2003.

[Biswas 04] Sanjit Biswas & Robert Morris. Opportunistic routing inmulti-hop wireless networks. ACM SIGCOMM ComputerCommunication Review, vol. 34, no. 1, page 69, janvier 2004.

[Biswas 05] Sanjit Biswas & Robert Morris. ExOR: opportunistic multi-hop routing for wireless networks. ACM SIGCOMM ComputerCommunication Review, vol. 35, no. 4, page 144, 2005.

[Bletsas 06] A. Bletsas, A. Khisti, D.P. Reed & A. Lippman. A simpleCooperative diversity method based on network path selection.IEEE Journal on Selected Areas in Communications, vol. 24,no. 3, pages 659 – 672, 2006.

[Blum 03] B Blum, T. He, S. Son & J. Stankovic. IGF: A state-free robustcommunication protocol for wireless sensor networks. In Tech.rep. CS-2003-11. Department of Computer Science, Universityof Virginia., 2003.

[Bose 01] Prosenjit Bose, Pat Morin, Ivan Stojmenovic & Jorge Urrutia.Routing with guaranteed delivery in ad hoc wireless networks.Wireless. Netw., vol. 7, no. 6, pages 609–616, 2001.

[Brennan 03] D.G. Brennan. Linear diversity combining techniques. IEEEProceedings, vol. 91, no. 2, pages 331–356, Feb. 2003.

[Cao 04] Qing Cao & T. Abdelzaher. A Scalable Logical CoordinatesFramework for Routing in Wireless Sensor Networks. 25thIEEE International Real-Time Systems Symposium, vol. V,pages 349–358, 2004.

[Caruso 05] Antonio Caruso, Stefano Chessa, Swades De & AlessandroUrpi. GPS free coordinate assignment and routing in wire-less sensor networks. In Proceedings IEEE 24th Annual JointConference of the IEEE Computer and Communications Soci-eties., volume 1, pages 150–160. IEEE, 2005.

[Coronel 07] Pedro Coronel, Robin Doss & Wolfgang Schott. GeographicRouting with Cooperative Relaying and Leapfrogging in Wire-less Sensor Networks. Communications Society, pages 646–651, 2007.

[Cover 79] T. Cover & A.E. Gamal. Capacity theorems for the relay chan-nel. Information Theory, IEEE Transactions on, vol. 25, no. 5,pages 572 – 584, sep 1979.

Bibliography 143

[Finn 87] G.G. Finn & UNIVERSITY OF SOUTHERN CALIFOR-NIA MARINA DEL REY INFORMATION SCIENCES INST.Routing and addressing problems in large metropolitan-scaleinternetworks. 1987.

[Fonseca 05] Rodrigo Fonseca, Sylvia Ratnasamy, Jerry Zhao, CT Ee,David Culler, Scott Shenker & I. Stoica. Beacon vector routing:Scalable point-to-point routing in wireless sensornets. In Pro-ceedings of the 2nd conference on Symposium on NetworkedSystems Design & Implementation, pages 329–342, 2005.

[Fubler 03] Holger Fubler, Joumlrg Widmer, Michael Kaumlsemann, Mar-tin Mauve & Hannes Hartenstein. Contention-based forward-ing for mobile ad hoc networks. Ad Hoc Networks, vol. 1, no. 4,page 18, Nov 2003.

[Gao 05] Jie Gao, L.J. Guibas, J. Hershberger, Li Zhang & An Zhu.Geometric spanners for routing in mobile networks. SelectedAreas in Communications, IEEE Journal on, vol. 23, no. 1,pages 174 – 185, jan. 2005.

[Haas 97] Zygmunt Haas. A new routing protocol for reconfigurable wire-less networks. In Proceedings of the IEEE International Con-ference on Universal Personal Communications, pages 562–566, October 1997.

[Heissenbuttel 03] Marc Heissenbuttel, Torsten Braun, Thomas Bernoulli &Markus Walchli. BLR: Beacon-Less Routing Algorithm forMobile Ad-Hoc Networks. Computer Communications, vol. 27,no. 11, pages 1076–1086, 2003.

[Herhold 04] P. Herhold, E. Zimmermann & G. Fettweis. A simple coopera-tive extension to wireless relaying. In Proc. Int. Zurich SeminarCommun., pages 36–39, Feb. 2004.

[Ibrahim 08a] Ahmed S Ibrahim, Zhu Han & KJR Liu. Distributed energy-efficient cooperative routing in wireless networks. IEEE Trans-actions on Wireless Communications, vol. 7, no. 10, pages3930–3941, 2008.

[Ibrahim 08b] A.S. Ibrahim, A.K. Sadek, Weifeng Su & K.J. Liu. Cooperativecommunications with relay-selection: when to cooperate andwhom to cooperate with? IEEE Trans. Wirel. Commun., vol. 7,pages 2814–2827, Jul 2008.

[Johnson 96] David B. Johnson & David A. Maltz. Dynamic Source Routingin Ad Hoc Wireless Networks. In Mobile Computing, pages153–181. Kluwer Academic Publishers, 1996.

144 Bibliography

[Johnson 04] David B. Johnson, David A. Maltz & Yih-Chun Hu. Thedynamic source routing protocol for mobile ad hoc networks(DSR), July 2004. Internet Draft draft-ietf-manet-dsr-10, In-ternet Engineering Task Force, MANET working group.

[Kalosha 08] Hanna Kalosha, Amiya Nayak, Stefan Ruhrup & Ivan Stoj-menovic. Select-and-Protest-Based Beaconless Georouting withGuaranteed Delivery in Wireless Sensor Networks. In INFO-COM 2008: The 27th Conference on Computer Communica-tions. IEEE, pages 346–350, April 2008.

[Karp 00] Brad Karp & H. T. Kung. GPSR: greedy perimeter statelessrouting for wireless networks. In MobiCom ’00: Proceedings ofthe 6th annual international conference on Mobile computingand networking, pages 243–254, 2000.

[Khandani 07] Amir Ehsan Khandani, Jinane Abounadi, Eytan Modiano &Lizhong Zheng. Cooperative Routing in Static Wireless Net-works. IEEE Transactions on Communications, vol. 55, no. 11,pages 2185–2192, novembre 2007.

[Ko 00] Young-Bae Ko & Nitin H. Vaidya. Location-aided routing(LAR) in mobile ad hoc networks. Wirel. Netw., vol. 6, no. 4,pages 307–321, 2000.

[Koetter 02] R. Koetter & M. Medard. Beyond Routing: An AlgebraicApproach to Network Coding. In INFOCOM, pages 122 – 130vol.1, 2002.

[Kuhn 02] Fabian Kuhn, Roger Wattenhofer & Aaron Zollinger. Asymp-totically optimal geometric mobile ad-hoc routing. In DIALM’02: Proceedings of the 6th international workshop on Discretealgorithms and methods for mobile computing and communi-cations, pages 24–33, New York, NY, USA, 2002. ACM.

[Kuhn 03a] Fabian Kuhn, Roger Wattenhofer, Yan Zhang & AaronZollinger. Geometric ad hoc routing: Of theory and practice.In PODC ’03: Proceedings of the twenty-second ACM sym-posium on Principles of distributed computing, pages 63–72,2003.

[Kuhn 03b] Fabian Kuhn, Roger Wattenhofer & Aaron Zollinger. Worst-Case optimal and average-case efficient geometric ad-hoc rout-ing. In MobiHoc ’03: Proceedings of the 4th ACM interna-tional symposium on Mobile ad hoc networking & computing,pages 267–278, 2003.

Bibliography 145

[Lai 06] Lifeng Lai, Ke Liu & H. El Gamal. The three-node wirelessnetwork: achievable rates and Cooperation strategies. Infor-mation Theory, IEEE Transactions on, vol. 52, no. 3, pages805 –828, march 2006.

[Laneman 03] J.N. Laneman & G.W. Wornell. Distributed space-time-codedprotocols for exploiting cooperative diversity in wireless net-works. IEEE Transactions on Information Theory, vol. 49,no. 10, pages 2415–2425, octobre 2003.

[Laneman 04] J.N. Laneman, D.N.C. Tse & G.W. Wornell. Cooperative di-versity in wireless networks: Efficient protocols and outage be-havior. IEEE Transactions on Information Theory, vol. 50,no. 12, pages 3062–3080, octobre 2004.

[Langendoen 05] K. Langendoen & N. Reijers. Distributed Localization Algo-rithms. In R. Zurawski, editeur, Embedded Systems Hand-book, pages 36.1 – 36.23. CRC press, 2005.

[Leong 06] Ben Leong, Barbara Liskov & Robert Morris. Geographic rout-ing without planarization. In NSDI’06: Proceedings of the 3rdconference on Networked Systems Design & Implementation,2006.

[Leong 07] Ben Leong, Barbara Liskov & Robert Morris. Greedy VirtualCoordinates for Geographic Routing. 2007 IEEE InternationalConference on Network Protocols, pages 71–80, octobre 2007.

[Li 02] Xiang-Yang Li, G. Calinescu & Peng-Jun Wan. Distributedconstruction of a planar spanner and routing for ad hoc wire-less networks. In INFOCOM 2002. Twenty-First Annual JointConference of the IEEE Computer and Communications Soci-eties. Proceedings. IEEE, volume 3, pages 1268 – 1277 vol.3,2002.

[Li 03] S.-Y.R. Li, R.W. Yeung & Ning Cai. Linear network cod-ing. Information Theory, IEEE Transactions on, vol. 49, no. 2,pages 371 –381, feb. 2003.

[Li 06] F. Li, K. Wu & A. Lippman. Energy-efficient cooperative rout-ing in multi-hop wireless ad hoc networks. In Proc. IEEEInternational Performance, Computing, and CommunicationsConference, pages 215–222, April 2006.

[Liu 05] Pei Liu, Zhifeng Tao & S. Panwar. A cooperative MAC protocolfor wireless local area networks. In IEEE International Con-ference on Communications, 2005. ICC 2005. 2005, volume 5,pages 2962–2968. IEEE, 2005.

146 Bibliography

[Liu 07a] Pei Liu, Zhifeng Tao, Sathya Narayanan, Thanasis Korakis &Shivendra Panwar. CoopMAC: A Cooperative MAC for Wire-less LANs. IEEE Journal on Selected Areas in Communica-tions, vol. 25, no. 2, pages 340–354, february 2007.

[Liu 07b] Shengpu Liu & Liang Cheng. Local Tree Based GeometricRouting. In Communications, 2007. ICC ’07. IEEE Interna-tional Conference on, pages 5478 –5483, 24-28 2007.

[Liu 08] Ke Liu & Nael Abu-ghazaleh. Aligned Virtual Coordinatesfor Greedy Geometric Routing in Wireless Sensor Networks.International Journal of Sensor Networks, vol. 3, no. 4, pages1–18, 2008.

[Mainaud 08] Bastien Mainaud, Vincent Gauthier & Hossam Afifi. Cooper-ative communication for Wireless Sensors Network : A Macprotocol solution. In Wireless Days, 2008. WD ’08. 1st IFIP,pages 1–5. IEEE, novembre 2008.

[Matlab ] Matlab. http://www.mathworks.com.

[MiXiM ] MiXiM. Mixed Simulator, http://mixim.sourceforge.net.

[Nelson 84] R. Nelson & L. Kleinrock. The Spatial Capacity of a SlottedALOHA Multihop Packet Radio Network with Capture. Com-munications, IEEE Transactions on, vol. 32, no. 6, pages 684– 694, jun 1984.

[Networkx ] Networkx. http://www.networkx.lanl.gov.

[Newsome 04] James Newsome & Dawn Song. GEM: Graph EMbedding forRouting and Data-Centric Storage in Sensor Networks With-out Geographic Information. 2004.

[Niculescu 04] D. Niculescu & B. Nath. Position and orientation in ad hocnetworks. Ad hoc networks, vol. 2, no. 2, pages 133–151, 2004.

[Ochiai 05] H. Ochiai, P. Mitran, H.V. Poor & V. Tarokh. Collaborativebeamforming for distributed wireless ad hoc sensor networks.IEEE Trans. Signal Processing, vol. 53, no. 11, pages 4110–4124, Nov. 2005.

[Ogier 04] R. Ogier, F. Templin & M. Lewis. Topology DisseminationBased on Reverse-Path Forwarding (TBRPF), 2004.

[Omnetpp ] Omnetpp. Even-driven simulator, http://www.omnetpp.org.

[Perkins 94] Charles E. Perkins & Pravin Bhagwat. Highly dynamicDestination-Sequenced Distance-Vector routing (DSDV) formobile computers. SIGCOMM Comput. Commun. Rev.,vol. 24, no. 4, pages 234–244, 1994.

Bibliography 147

[Perkins 03] C. Perkins, E. Belding-Royer & S. Das. Ad hoc On-DemandDistance Vector (AODV) Routing, 2003.

[Python ] Python. http://www.python.org.

[Rao 03] Ananth Rao, Christos Papadimitriou, Scott Shenker & IonStoica. Geographic routing without location information. InProceedings of the 9th annual international conference on Mo-bile computing and networking - MobiCom ’03, page 96, NewYork, New York, USA, 2003. ACM Press.

[RFC 03] Optimized Link State Routing Protocol (OLSR), 2003.

[Sadek 07] A. K. Sadek, W. Su & K. J. R. Liu. Multinode cooperativecommunications in wireless networks. IEEE Trans. Signal Pro-cessing, vol. 55, no. 1, pages 341–355, Jan. 2007.

[Sadek 08] A. K. Sadek, W. Su & K. J. R. Liu. Cooperative commu-nication protocols in wireless networks: performance analysisand optimum power allocation. Wirel. Pers. Commun., vol. 44,pages 181–217, Jan. 2008.

[Sanchez 07] Juan A. Sanchez, Rafael Marin-Perez & Pedro M. Ruiz. BOSS:Beacon-less On Demand Strategy for Geographic Routing in-Wireless Sensor Networks. In MASS 2007: The 4th Interna-tonal Conference on Mobile Ad-hoc and Sensor Systems, pages1–10, 2007.

[Sanchez 09] J. A. Sanchez, P. M. Ruiz & R. Marin-Perez. Beacon-less geo-graphic routing made practical: challenges, design guidelines,and protocols. IEEE Commun. Magazine, vol. 47, no. 8, pages85–91, Aug. 2009.

[Santi 03] Paolo Santi & Douglas M. Blough. The Critical TransmittingRange for Connectivity in Sparse Wireless Ad Hoc Networks.IEEE Transactions on Mobile Computing, vol. 2, no. 1, pages25–39, 2003.

[Savarese 01] C. Savarese, JM Rabaey & J. Beutel. Location in distributedad-hoc wireless sensor networks. In 2001 IEEE InternationalConference on Acoustics, Speech, and Signal Processing, 2001.Proceedings.(ICASSP’01), volume 4, 2001.

[Seddigh 01] Mahtab Seddigh, Julio Solano Gonzalez & Ivan Stojmenovic.RNG and internal node based broadcasting algorithms for wire-less one-to-one networks. SIGMOBILE Mob. Comput. Com-mun. Rev., vol. 5, no. 2, pages 37–44, 2001.

148 Bibliography

[Sendonaris 03] A. Sendonaris, E. Erkip & B. Aazhang. User cooperationdiversity-Part I,II. IEEE Trans. Commun., vol. 51, pages1927–1938, Nov 2003.

[Shan 08] H.-S. Shan, W. Wang, W. Zhuang & Z. Wang. Cross-Layer Co-operative Triple Busy Tone Multiple Access for Wireless Net-works. In Proc. 2008 IEEE Globecom, pages 1–5, 2008.

[Simon 98] M.K. Simon & M. Alouini. A unified approach to the per-formance analysis of digital communication over generalizedfading channels. IEEE Proceedings, vol. 86, no. 9, pages 1860–1877, Sept. 1998.

[Singh 99] H. Singh & J. Urrutia. Compass Routing on Geometric Net-works. In Proc. of the 11th Canadian Conference on Compu-tational Geometry, Vancouver, Canada, August 1999.

[Stanojev 06] I. Stanojev, O. Simeone, Y. Bar-Ness & C. You. Performanceof multi-relay collaborative hybrid-ARQ protocols over fadingchannels. Communications Letters, IEEE, vol. 10, no. 7, pages522 –524, july 2006.

[Stojmenovic 02] I. Stojmenovic. Position-based routing in ad hoc network.IEEE Commun. Mag., vol. 40, no. 7, pages 128–134, Jul 2002.

[Su 08] Weifeng Su, Ahmed K. Sadek & K. J. Ray Liu. CooperativeCommunication Protocols in Wireless Networks: PerformanceAnalysis and Optimum Power Allocation. Wireless PersonalCommunications, vol. 44, no. 2, pages 181–217, 2008.

[Takagi 84] H. Takagi & L. Kleinrock. Optimal Transmission Ranges forRandomly Distributed Packet Radio Terminals. Communica-tions, IEEE Transactions on, vol. 32, no. 3, pages 246–257,March 1984.

[Toussaint 80] Godfried T. Toussaint. The Relative Neighbourhood Graph ofa Finite Planar Set. Pattern Recognition, vol. 12, pages 261–268, 1980.

[Valenti 04] M.C. Valenti & Bin Zhao. Hybrid-ARQ based intra-clustergeographic relaying. IEEE MILCOM 2004. Military Commu-nications Conference, 2004., pages 805–811, 2004.

[Vardhe 10] K. Vardhe, D. Reynolds & B.D. Woerner. Joint power allo-cation and relay selection for multiuser cooperative commu-nication. IEEE Trans. Wirel. Commun., vol. 9, no. 4, pages1255–1260, Apr. 2010.

[Wang 09a] C.-L. Wang & S.-J. Syue. An efficient relay selection protocolin cooperative wireless sensor networks. In Proc. 2009 IEEE

Bibliography 149

Wireless Commun. and Networking Conf. (WCNC 2009),pages 1–5, Budapest, Hungary, April 2009.

[Wang 09b] C.-L. Wang & S.-J. Syue. An efficient relay selection protocolin cooperative wireless sensor networks. In Proc. 2009 IEEEWireless Commun. and Networking Conf. (WCNC 2009),pages 1–5, Budapest, Hungary, April 2009.

[Wang 09c] Chin-Liang Wang & Syue-Ju Syue. A Geographic-Based Ap-proach to Relay Selection in Wireless Ad Hoc Relay Networks.In Proc. 69th Vehicular Technology Conference. VTC Spring2009. IEEE, April 2009.

[Watteyne 06] Thomas Watteyne, Abdelmalik Bachir, Mischa Dohler, Do-minique Barthe & Isabelle Auge-Blum. 1-hopMAC: AnEnergy-Efficient MAC Protocol for Avoiding 1 -hop Neighbor-hood Knowledge. In 2006 3rd Annual IEEE CommunicationsSociety on Sensor and Ad Hoc Communications and Networks,volume 2, pages 639–644. IEEE, septembre 2006.

[Watteyne 07] Thomas Watteyne, David Simplot-Ryl, Isabelle Auge-Blum &Mischa Dohler. On using Virtual Coordinates for Routing inthe Context of Wireless Sensor Networks. 2007 IEEE 18th In-ternational Symposium on Personal, Indoor and Mobile RadioCommunications, pages 1–5, septembre 2007.

[Watteyne 09] Thomas Watteyne, Isabelle Auge-Blum, Mischa Dohler,Stephane Ubeda & Dominique Barthel. Centroid virtual co-ordinates A novel near-shortest path routing paradigm. Com-puter Networks, vol. 53, no. 10, pages 1697–1711, 2009.

[Yao 82] Andrew Chi-Chih Yao. On Constructing Minimum SpanningTrees in k-Dimensional Spaces and Related Problems. SIAMJournal on Computing, vol. 11, no. 4, pages 721–736, 1982.

[Yi 08] Z. Yi & I.-M. Kim. Diversity order analysis of the decode-and-forward cooperative networks with relay selection. IEEE Trans.Wirel. Commun., vol. 7, no. 5, pages 1792–1799, May 2008.

[Zhao 05a] B Zhao & MC Valenti. Practical relay networks: a general-ization of hybrid-ARQ. IEEE Journal on Selected Areas inCommunications, vol. 23, no. 1, pages 7–18, 2005.

[Zhao 05b] Yao Zhao, Bo Li, Qian Zhang, Yan Chen & Wenwu Zhu. Ef-ficient Hop ID based Routing for Sparse Ad Hoc Networks.13TH IEEE International Conference on Network Protocols(ICNP’05), pages 179–190, 2005.

150 Bibliography

[Zhao 07] Yi Zhao, Raviraj Adve & Teng Joon Lim. Improving amplify-and-forward relay networks: optimal power allocation versusselection. IEEE Transactions on Wireless Communications,vol. 6, no. 8, pages 3114–3123, 2007.

[Zhong 09] Z. Zhong, Shihua Zhu & A. allanathan. Delay-tolerant dis-tributed linear convolutional space-time code with minimummemory length under frequency-selective channels. IEEETrans. Wirel. Commun., vol. 8, no. 8, pages 3944–3949, Aug.2009.

[Zorzi 03] Michele Zorzi & Ramesh R. Rao. Geographic random forward-ing (GeRaF) for ad hoc and sensor networks: Multihop per-formance. IEEE Transactions on Mobile Computing, vol. 2,no. 4, pages 337–348, 2003.

Glossary

BER Bit Error Rate, 41

BFP Beaconless Forwarder Planarization, 69

BLGF BeaconLess Greedy Forwarding, 66

BLGR BeaconLess Geographic Routing, 61

BLR BeaconLess Routing, 31

BLRF BeaconLess Recovery Forwarding, 66

BOSS Beaconless On Demand Strategy for Geographic Routing in WirelessSensor Networks, 36

CBF Contention-Based Forwarding, 31

CoopGeo Beaconless Cooperative Geographic cross-layer protocol, 9

CSA Common Sub-Areas, 67

CSI Channel State Information, 44

CTF Clear-To-Forward, 67

GeRaF Geographic Random Forwarding, 35

GFG Greedy Face Greedy, 28

GPSR Greedy Perimeter Stateless Routing, 28

GR Geographic Routing, 23

IGF Implicit Geographic Routing, 31

MIMO Multiple-Input Multiple-Output, 60

MQAM M-Quadrature Amplitude Modulation, 79

MRC Maximum Ratio Combining, 41

151

152 Bibliography

NPA Negative Progress Area, 66

PER Packet Error Rate, 77

PPA Positive Progress Area, 66

RACR Relay-Aware Cooperative Routing protocol, 9

SER Symbol Error Rate, 9, 20

SNR Signal to Noise Ratio, 44

List of Figures

2.1 Greedy Forwarding: Node s forwards the packet to neighbor F1 . 262.2 Forwarding strategies . . . . . . . . . . . . . . . . . . . . . . . . . 282.3 Nearest and farthest neighbor strategies . . . . . . . . . . . . . . 282.4 In blue the right hand rule and in red the face changes, two princi-

ples composing the face traversal on a planar graph used in GFGand GPSR algorithms strategies . . . . . . . . . . . . . . . . . . . 30

2.5 60 sector from S to D within the transmission range . . . . . . . 332.6 Area-based suppression strategies. Node A,B and C are in the

reuleaux triangle whereas only node C is in the circle area . . . . 352.7 Forwarding area sequence shifts in IGF . . . . . . . . . . . . . . . 362.8 Regions from the destination point of view . . . . . . . . . . . . . 372.9 BOSS extends the positive sub-regions from GeRaf to the negative

area of the node coverage area . . . . . . . . . . . . . . . . . . . . 382.10 The node P builds its virtual coordinates vector using the three

landmark nodes (yellow) . . . . . . . . . . . . . . . . . . . . . . . 412.11 In phase 1, the source broadcast its data and relay and destination

receive it. In phase 2, the relay node retransmit the data receivedin phase 1. In these two phases, three different fading paths are used 44

3.1 Unit Disk Graph . . . . . . . . . . . . . . . . . . . . . . . . . . . 523.2 Unit Disk Graph network . . . . . . . . . . . . . . . . . . . . . . 523.3 Gabriel Graph model and network . . . . . . . . . . . . . . . . . . 533.4 RNG Graph model and network . . . . . . . . . . . . . . . . . . . 543.5 Delaunay triangulation model and Delaunay network . . . . . . . 553.6 The Random Waypoint Model behavior, where each color repre-

sents the movement of a node . . . . . . . . . . . . . . . . . . . . 603.7 Brownian motion model in a single node . . . . . . . . . . . . . . 60

153

154 List of Figures

4.1 (a) Cooperative multihop sensor network model (b) Direct andcooperative modes for each hop . . . . . . . . . . . . . . . . . . . 67

4.2 Area division for CoopGeo routing. F1 and F2 are sub-area 0 and1 of PPA respectively, whereas F3 and F4 are sub-area 4 and 5 ofNPA respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.3 CoopGeo architecture . . . . . . . . . . . . . . . . . . . . . . . . . 714.4 Recovery forwarding area is divided in N coronas. Each has a

width (√i−√i− 1)r1 . . . . . . . . . . . . . . . . . . . . . . . . 74

4.5 Nodes vi...vn have the same distance to u. So, each node has to usea random function rand(Tmax

NSA) to decrease the collision probability 75

4.6 Beaconless recovery messages exchange . . . . . . . . . . . . . . . 764.7 Beaconless Recovery Forwarding happens at NPA area when the

Beaconless Greedy Forwarding fails . . . . . . . . . . . . . . . . . 774.8 (a) Mapping of the metric on to the set C (b) Mapping of the metric

on to the set D for a normalized distance Source(0,0) Destination(1,0) 794.9 CoopGeo in action . . . . . . . . . . . . . . . . . . . . . . . . . . 804.10 Comparison of each possible relay selection and random relay se-

lection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 824.11 PER for BOSS and CoopGeo using Tmax = 100...1000 µs. The

curves located at the bottom of the figures correspond to minimumvalue of Tmax = 100 and those located in the upper side to themaximum value Tmax = 1000 . . . . . . . . . . . . . . . . . . . . . 84

4.12 Packet Error Rate for Tmax = 500µs . . . . . . . . . . . . . . . . 854.13 Error Transmission Rate (end to end) for BOSS and CoopGeo

using Tmax = [100,...,1000] µs . . . . . . . . . . . . . . . . . . . . 864.14 End to End Transmission Error Probability for Tmax = 500µs . . 874.15 CTF-Relayed message collision probability when changing Tmax

from 100µs to 1000µs . . . . . . . . . . . . . . . . . . . . . . . . . 884.16 Saturation throughput for QAM: 16-32 . . . . . . . . . . . . . . . 884.17 Saturation throughput for QAM: 64-128 . . . . . . . . . . . . . . 894.18 CoopGeo Saturated throughput for QAM fom 16-128 . . . . . . . 894.19 Normalized saturated throughput and collision probability for Tmax

= 300µs and Tmax = 500µs . . . . . . . . . . . . . . . . . . . . . 90

5.1 Extended coverage using cooperative transmission . . . . . . . . . 935.2 Multihop sensor network with cooperative geographic routing. . . 965.3 RACR Architecture. . . . . . . . . . . . . . . . . . . . . . . . . . 1005.4 Optimal relaying and forwarding positions and the direct and max-

imum cooperative transmission radii. . . . . . . . . . . . . . . . . 1015.5 Example of optimal relay and forwarding positions distributions . 102

List of Figures 155

5.6 Relay selection as function of alpha (a)Alpha: 2 (b)Alpha: 3c()Alpha: 4 (d)Alpha: 3.8 . . . . . . . . . . . . . . . . . . . . . . 104

5.7 Coverage extension (%) with alpha:4, due to cooperation versusthe relaying position with a cross-sectional view for (a) 4QAM,(b) 16QAM, (c)32QAM, (d)64QAM. . . . . . . . . . . . . . . . . 105

5.8 Coverage extension (%) with alpha:4, due to cooperation versusthe relaying position with a 3D view for (a)4QAM, (b)16QAM,(c)32QAM, (d)64QAM. . . . . . . . . . . . . . . . . . . . . . . . . 106

5.9 (a)Average path length versus the average number of neighbors.(b)The corresponding stretch factor. . . . . . . . . . . . . . . . . 107

C.1 (a) Cooperative multihop sensor network model (b) Direct andcooperative modes for each hop . . . . . . . . . . . . . . . . . . . 122

C.2 Area division for CoopGeo routing. F1 and F2 are sub-area 0 and1 of PPA respectively, whereas F3 and F4 are sub-area 4 and 5 ofNPA respectively. . . . . . . . . . . . . . . . . . . . . . . . . . . 124

C.3 PER de BOSS et CoopGeo avec Tmax = 100...1000 µs. Les courbessituees dans la partie inferieure du graph correspondent a Tmax =100 et celles situees dans la partie superieure correspondent a lavaleur maximale Tmax = 1000 . . . . . . . . . . . . . . . . . . . . 129

C.4 Packet Error Rate for Tmax = 500µs . . . . . . . . . . . . . . . . 130C.5 End to End Transmission Error Probability for Tmax = 500µs . . 131C.6 (a)CTF-Relayed message collision probability when changing Tmax

from 100µs to 1000µs. (b)CoopGeo Saturated throughput for QAMfom 16-128. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 132

C.7 Positions optimales du relais et intermediaire, et rayon de trans-mission direct et cooperatif. . . . . . . . . . . . . . . . . . . . . . 136

C.8 Coverage extension (%) with alpha :4, due to cooperation versusthe relaying position with a 3D view for (a)4QAM, (b)16QAM,(c)32QAM, (d)64QAM. . . . . . . . . . . . . . . . . . . . . . . . . 138

C.9 (a)Average path length versus the average number of neighbors.(b)The corresponding stretch factor. . . . . . . . . . . . . . . . . 139

List of Tables

4.1 Relays Locations and The Corresponding Selection Metrics . . . . 824.2 Simulation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . 83

5.1 Simulation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . 107

C.1 Simulation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . 128C.2 Simulation Settings . . . . . . . . . . . . . . . . . . . . . . . . . . 139

157

ecole doctorale edite thèse présentée pour l ... · ecole doctorale edite thèse présentée...

Documents